econstor
A Service of
zbw
Make Your Publications Visible.
Leibniz-Informationszentrum
Wirtschaft
Leibniz Information Centre
for Economics
Dachs, Bernhard; Hud, Martin; Koehler, Christian; Peters, Bettina
Working Paper
Employment effects of innovations over the business
cycle: Firm-level evidence from European countries
ZEW Discussion Papers, No. 16-076
Provided in Cooperation with:
ZEW - Zentrum für Europäische Wirtschaftsforschung / Center for
European Economic Research
Suggested Citation: Dachs, Bernhard; Hud, Martin; Koehler, Christian; Peters, Bettina (2016) :
Employment effects of innovations over the business cycle: Firm-level evidence from European
countries, ZEW Discussion Papers, No. 16-076
This Version is available at:
http://hdl.handle.net/10419/148057
Standard-Nutzungsbedingungen:
Terms of use:
Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen
Zwecken und zum Privatgebrauch gespeichert und kopiert werden.
Documents in EconStor may be saved and copied for your
personal and scholarly purposes.
Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich
machen, vertreiben oder anderweitig nutzen.
You are not to copy documents for public or commercial
purposes, to exhibit the documents publicly, to make them
publicly available on the internet, or to distribute or otherwise
use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten,
gelten abweichend von diesen Nutzungsbedingungen die in der dort
genannten Lizenz gewährten Nutzungsrechte.
www.econstor.eu
If the documents have been made available under an Open
Content Licence (especially Creative Commons Licences), you
may exercise further usage rights as specified in the indicated
licence.
Dis­­cus­­si­­on Paper No. 16-076
Employment Effects of Innovations
over the Business Cycle:
Firm-Level Evidence from
European Countries
Bernhard Dachs, Martin Hud, Christian Koehler,
and Bettina Peters
Dis­­cus­­si­­on Paper No. 16-076
Employment Effects of Innovations
over the Business Cycle:
Firm-Level Evidence from
European Countries
Bernhard Dachs, Martin Hud, Christian Koehler,
and Bettina Peters
Download this ZEW Discussion Paper from our ftp server:
http://ftp.zew.de/pub/zew-docs/dp/dp16076.pdf
Die Dis­­cus­­si­­on Pape­rs die­­nen einer mög­­lichst schnel­­len Ver­­brei­­tung von
neue­­ren For­­schungs­­arbei­­ten des ZEW. Die Bei­­trä­­ge lie­­gen in allei­­ni­­ger Ver­­ant­­wor­­tung
der Auto­­ren und stel­­len nicht not­­wen­­di­­ger­­wei­­se die Mei­­nung des ZEW dar.
Dis­­cus­­si­­on Papers are inten­­ded to make results of ZEW ­research prompt­­ly avai­­la­­ble to other
eco­­no­­mists in order to encou­­ra­­ge dis­­cus­­si­­on and sug­­gesti­­ons for revi­­si­­ons. The aut­­hors are sole­­ly
respon­­si­­ble for the con­­tents which do not neces­­sa­­ri­­ly repre­­sent the opi­­ni­­on of the ZEW.
Employment Effects of Innovations over the Business
Cycle: Firm-Level Evidence from European Countries
Bernhard Dachs1
Martin Hud2,3
1
2
Christian Koehler2
Bettina Peters2,4,5
AIT Austrian Institute of Technology, Vienna, Austria
ZEW Centre for European Economic Research, Mannheim, Germany
3
4
Maastricht University, Netherlands
Mannheim Centre for Competition and Innovation
5
University of Luxembourg, Luxembourg
November 2016
Abstract
A growing literature investigates how firms’ innovation input reacts to changes in the business cycle.
However, so far there is no evidence whether there is cyclicality in the effects of innovation on firm
performance as well. In this paper, we investigate the employment effects of innovations over the
business cycle. Our analysis employs a large data set of manufacturing firms from 26 European countries
over the period from 1998 to 2010. Using the structural model of Harrison et al. (2014), our empirical
analysis reveals four important findings: First, the net effect of product innovation on employment growth
is pro-cyclical. It turns out to be positive in all business cycle phases except for the recession. Second,
product innovators are more resilient to recessions than non-product innovators. Even during recessions
they are able to substitute demand losses from old products by demand gains of new products to a
substantial degree. As a result their net employment losses are significantly lower in recessions than those
of non-product innovators. Third, we only find resilience for SMEs but not for large firms. Fourth,
process and organizational innovations displace labor primarily during upturn and downturn periods.
JEL: O33, J23, C26, D2
Keywords: Innovation, employment, business cycle, resilience, Europe
Acknowledgements: We would like to thank Giuliana Battisti, Martin Carree, Claudio Fassio, Aldo Geuna, Jordi
Jaumandreu, Cindy Lopes Bento, Jacques Mairesse, Christian Rammer, Tom Schmitz, Konrad Stahl and Marco
Vivarelli for their valuable comments. Furthermore, we thank participants at the ZEW industrial economics lunch
seminar, PhD Workshop 2014 jointly organized by KU Leuven, University of Turin, Copenhagen Business School
and ZEW, EMAEE 2015, DRUID 2015, NoeG 2015, EARIE 2015, Atlanta Conference on Science and Innovation
Policy 2015, APIC 2015, IIOC 2016 and MEIDE 2016. In addition, we are grateful to Geneviève Villette at
Eurostat, Luxembourg, for providing data access and her support. This research is based on a project done for the
European Commission under the Framework Contract ENTR/2009/033. The opinions expressed in this study are
those of the authors and do not represent the European Commission’s official position. Any errors remain our own.
Contact information: Bernhard Dachs: [email protected]; Martin Hud: [email protected]; Christian Köhler:
[email protected]; Bettina Peters: [email protected].
1. Introduction
The global economic crisis set off in 2008 has been a serious threat to the stability of most economies
in the world. More than 5.37 million jobs (-2.4%) were destroyed between 2008 and 2010 just in
Europe. The challenge European policymakers face has been to overcome the crisis and to improve
long-term competitiveness and to stimulate growth. Research and development (R&D) and innovation
activities are typically regarded as efficient instruments to spur firms’ competitiveness and,
consequently, economic growth and job creation. For this reason, improving the conditions for
research and innovation is one of the main objectives of Europe 2020 – EU’s large-scale growth
strategy implemented in 2010 (EC, 2012). The key issue whether such a strategy can be successful
depends on the extent to which EU countries are able to translate new products and technologies into
employment growth during a recession.
In this paper, we analyze the cyclicality of the link between innovation and job creation for a large set
of European firms. Our central research question is: How do different types of innovation affect
employment growth during different phases of the business cycle? In this sense, is innovation equally
employment-creating in all phases of the business cycle or do we observe a pro- or counter-cyclical
effect of innovation on employment growth? Investigating employment creation and destruction of
innovation over the business cycle, allows us to answer in particular the intriguing research question
whether or not economic activities of innovators are more resilient to negative economic shocks 1. The
answers largely rely on two different effects. The first effect refers to the dependency of firms’
innovation activities on business cycle fluctuations. A growing literature has studied firms’ innovation
input behavior over the business cycle and has found that firms expand these activities during
economic upturns (see e.g. Barlevy, 2007; Fabrizio and Tsolmon, 2014). The second effect describes
the impact of innovation output on firms’ employment growth. The literature has shown that this effect
depends on several mechanisms in a complex manner. A main conclusion that can be drawn is that the
effect is considerably driven not only by the specific type of innovation but also by the demand for
firms’ products (see Pianta, 2005 for an overview). Product innovations have mainly been found to
stimulate firms’ labor demand, whereas the effect of process and organizational innovations is
ambiguous. Even though the literature stresses that demand-driven effects are crucial for employment
consequences of innovation, there is no firm-level study yet investigating the firms’ ability to
transform innovation into employment growth over the business cycle. We contribute to the literature
by providing first evidence regarding firm-level employment growth effects of innovation over
different phases of a business cycle.
1
Regional-industry level evidence points towards this direction. Delgado et al. (2015) found that strong regional clusters in
terms of patenting have facilitated employment resilience in the US during the 2007-2009 recession.
1
Our analysis disentangles this complex relationship by using the structural model of Harrison et al.
(2014). This model is conceived to examine the labor-creating and labor-destructing effects of
innovations. In this respect, the model establishes a theoretical link between firm-level employment
growth and innovation output in terms of (i) the sales growth generated by product innovations, i.e.
new or improved products, and (ii) the efficiency gains attributable to process innovations, i.e. new or
improved processes. We extend the model’s standard specification by allowing efficiency
improvements to depend on organizational innovations as well. Schumpeter (1934) already
emphasized that firms implement new organizational structures from time to time and do not only rely
on product and process innovations. Controlling for organizational innovations also allows us to
identify the employment growth effects of process innovations more accurately.
To estimate the model, we use data from the Europe-wide Community Innovation Survey (CIS). Our
sample covers information on manufacturing firms from 26 European countries and includes more
than 200,000 firm-year observations. We observe them for the period from 1998 to 2010, which is
reasonably long and allows us to capture potential business cycle effects. The estimations are
weighted, i.e. our findings are representative for the manufacturing sectors of the countries covered.
To analyze the cyclicality of the employment growth effects of innovation, we use country-specific
GDP growth rates to create dummy variables dividing the business cycle into four different phases, i.e.
upturn, boom, downturn and recession. Furthermore, we study whether or not employment creation
and destruction of innovation over the business cycle differs by firm size. Prior empirical evidence has
shown that jobs are typically created in small and medium-sized enterprises (SMEs) (see e.g. Neumark
et al., 2011).
Our empirical analysis reveals four important findings. First, the net employment growth effect of
product innovators is pro-cyclical. It turns out to be positive in all business cycle phases except for the
recession. This means, the potential labor-creating effect exceeds the potential labor-destructing effect
of product innovations except during recessions. Second, product innovators are more resilient to
recessions than firms that have not introduced product innovations (non-product innovators). When
facing negative economic growth, product innovators on average cut jobs. However, the level of job
destruction is much more modest as compared to the job destruction of non-product innovators. Third,
this resilience of product innovators is, however, only found in SMEs but not in large firms. Fourth, on
average, process and organizational innovations particularly reduce labor demand of firms during
upturn and downturn periods. Overall, our results suggest that product innovators are an important
driving force for firm-level employment growth. They are particularly important for securing jobs
during recessions. In contrast, process and organizational innovations tend to displace employment.
The paper is structured as follows. The next section reviews the relevant literature and develops the
hypotheses, while Section 3 describes the underlying theoretical and econometric model. Section 4
explains the empirical implementation by discussing the data, descriptive statistics and estimation
approach. Section 5 presents the empirical results. Section 6 briefly summarizes and concludes.
2
2. Related literature and hypotheses
This section reviews the literature relevant to our analysis and develops hypotheses. First, we present
the basic findings on firms’ innovation activities and their business cycle dependency. Second, we
shortly describe the theory on the employment growth effects of different types of innovation and
present the main empirical findings. Third, based on the literature review we develop our hypotheses
on the employment effects of innovations over the business cycle.
2.1. Innovation activities and the business cycle
The literature on the relationship between the business cycle and innovation has mostly focused on
innovation input. During recessions, firms shift more resources to productivity-enhancing activities,
such as innovative investment. This counter-cyclical pattern arises because the opportunity costs of
long-term innovative investment are lower than short-time capital investment during recessions (see
e.g. Bean, 1990; Gali and Hammour, 1991; Aghion and Saint-Paul, 1998). In contrast to these
predictions, recent empirical studies find a pro-cyclical pattern of innovative investment. Some authors
attribute this result to the cash-flow dependency of financing innovation activities (see e.g.
Himmelberg and Petersen, 1994; Mulkay et al., 2001; Ouyang, 2011). Other authors argue that the
investment decision is strategically postponed to high demand periods to maximize the innovations’
expected profitability (see e.g. Barlevy, 2007). Aghion et al. (2010, 2012) find evidence for pro- and
counter-cyclicality. Accordingly, pro-cyclicality arises for financially constrained firms, whereas nonconstrained firms act counter-cyclically.
In addition, there is some research on the timing of the market introduction and commercialization of
innovations. Shleifer (1986), Francois and Lloyd-Ellis (2003) and Barlevy (2007) develop similar
models and argue that the innovators expect imitators to enter after the implementation of
innovations. 2 Anticipating this, the innovators postpone the product commercialization to periods of
high demand. Only this allows them to capture most of the profits. By using US firm-level data,
Fabrizio and Tsolmon’s (2014) empirical results support this pro-cyclicality. Furthermore, Axalorglou
(2003) uses industry-level data and also finds a positive relationship between growth and the
introduction of new products.
2.2. Employment effects of innovations
Technological progress may be labor-saving or labor-creating, and may or may not cause a change in
the skill composition (see Vivarelli, 2014 for an overview). In what follows, we neglect the skill
2
In their model, Francois and Lloyd-Ellis (2003) do not build on imitation. Instead, they assume that after the
implementation knowledge disseminates to the rivals. This eventually limits the innovators’ time as successful incumbents.
3
aspect because our data does not include information on skills but only allows us to study total
employment growth effects of innovations at the firm level. 3
Theoretical contributions do not provide clear-cut predictions on whether or not innovations create or
destroy jobs (see Petit, 1995; Blechinger et al., 1998 for overviews). Identifying the employment
growth impact rather requires separating the effects of product from the effects of process innovations.
At the firm level, product innovations may affect employment via three channels. First, introducing
new products on the market generates new demand and therefore increases labor demand (direct
demand effect of product innovation). The second channel is related to the relative productivity
between new and old products. If new products are produced more (less) efficiently than old products,
they will require less (more) input for a given output. This dampens (strengthens) the positive demand
effect, thus also employment growth (productivity effect of product innovation). The third channel
refers to an indirect demand effect of product innovation. According to this, new product demand may
replace the demand for the innovators’ old products to some degree. This “product cannibalization”
reduces labor demand related to the old products. In contrast, the innovators’ labor demand will
increase if new and old products complement each other. In this case, new product demand stimulates
old product demand. Therefore, at the firm level product innovations only unambiguously increase
firms’ employment levels in case the relationship is complementary. Further employment effects arise
at the macro and the sector level. 4 For instance, the increase in new product demand may come at the
expense of lower demand for rivals’ products (business stealing effect). This reduces the rivals’ labor
demand. However, product and labor demand of those competitors that offer products complementary
to the innovators’ new products eventually increase.
Despite ambiguous theoretical predictions, the majority of empirical studies find product innovations
to create jobs (see e.g. Entorf and Pohlmeier, 1990; Brouwer et al., 1993; König et al., 1995; Van
Reenen, 1997; Smolny, 2002; Garcia et al., 2004; Hall et al., 2008; Peters, 2008; Lachenmaier and
Rottmann, 2011; Dachs and Peters, 2014; Harrison et al., 2014). This implies that the employmentinducing effects outweigh the potential employment-reducing effects of product innovations.
However, only Garcia et al. (2004), Hall et al. (2008), Peters (2008), Dachs and Peters (2014) and
Harrison et al. (2014) explicitly disentangle the labor-creating from the labor-reducing effects of
product innovations.
In contrast to product innovations, the direct effect of process innovations is an increase in the
innovators’ production efficiency (productivity effect of process innovation). This type of efficiency
gain implies that the same level of output can be produced with less input, e.g. labor. Hence, the
3
Two strands of the literature center on the impact of innovation on skills: The literature on skill-biased technological change
(see e.g. Caroli and Van Reenen, 2001; Acemoglu, 2002; Bresnahan et al., 2002; Piva et al., 2005) and on routine-biased
technological change (see e.g. Autor et al., 2003 and Goos et al., 2014).
4
For empirical research, see e.g. Freeman et al. (1982), Vivarelli and Pianta (2000), Leitner et al. (2011), Pianta and
Lucchese (2012) and Damijan et al. (2014).
4
productivity effect of process innovation is likely to reduce the innovators’ demand for labor.
However, as efficiency improvements cause marginal production costs to decline, they open up
possibilities for price reductions. Lower prices stimulate the innovators’ product demand. In this way,
price cuts can alleviate the employment losses related to the productivity effect or even reverse them
(price effect of process innovations). The magnitude of the price effect is determined by the size of the
price reduction, the price elasticity of demand and the competitive environment, in particular
competitors’ reaction to price reductions.
Empirical results for the effect of process innovations are inconclusive. Mainly based on reduced form
regressions, Entorf and Pohlmeier (1990), Van Reenen (1997) and Hall et al. (2008) report no
significant effect of process innovations on employment, whereas König et al. (1995) and
Lachenmaier and Rottmann (2011) even find that process innovations increase the firms’ employment
level. Using the same structural approach as this paper, Peters (2008), Dachs and Peters (2014) and
Harrison et al. (2014) find evidence for a small negative gross effect of process innovation, i.e. process
innovations cut labor due to improved productivity. But the growth of demand for old products –
partly provoked by price reductions following the increase in efficiency – is strong enough to
compensate for it.
The majority of the employment studies focus on product and process innovations (technological
innovations). This is a significant drawback because analyzing employment effects of innovations also
requires the adoption of a non-technological perspective in the form of organizational innovations
(Edquist et al., 2001). Schumpeter (1934) already stated that firms not only implement new products
and processes, but also adjust their business practices and reorganize their organizational structures.
However, organizational innovations have largely been neglected for some time due to measurement
and definition problems (Lam, 2005; Armbruster et al., 2008). OECD and Eurostat (2005) provide the
first harmonized definition of organizational innovations and how to measure them in innovation
surveys. Evangelista and Vezzani (2011) analyze CIS data of several European countries and
descriptively show that there are more organizational innovators than product and process innovators.
By using a different data set covering information on a large set of European firms, Tether and Tajar’s
results (2008) disclose a very similar pattern.
There is no theoretical model explicitly considering employment effects of organizational innovations.
However, Ichniowski et al. (1996) review the literature on workplace innovations and performance.
They argue that these kinds of innovations should increase productivity, which should lead to lower
costs and higher product demand. Indeed, empirical research supports the productivity-enhancing
effect of organizational innovations (see e.g. Ichniowski et al., 1997; Black and Lynch, 2004; Gera and
Gu, 2004). Like in the case of process innovations, the evidence as to the employment growth effect of
organizational innovations is ambiguous as well. Greenan (2003) shows that a shift towards a flexible
enterprise increases the firms’ job destruction rate. Likewise, Bauer and Bender (2004) find delayering
and the transfer of responsibilities to significantly decrease net employment growth rates, whereas
5
team work causes the employment level to increase. Positive employment effects have also been
found, for instance, by Falk (2001), Addison et al. (2008) and Evangelista and Vezzani (2011).
2.3. Hypotheses: Employment effects of innovations over the business cycle
Based on the related literature presented in Section 2.2, it is evident that product demand plays an
important role for employment growth effects of technological and non-technological innovations.
These demand effects are likely to vary with different phases of a business cycle as do the productivity
effects of process and organizational innovations. In this section, we develop a set of hypotheses about
the employment growth effects of different types of innovations at different phases of the business
cycle.
The gross employment effect of product innovations depends on two effects: (i) the size of the direct
demand effect and (ii) the size and direction of the productivity effect. The latter refers to the
productivity of new products, i.e. the level of output per unit of input, relative to the productivity of
old products. As this is mainly technology-driven, we do not expect the relative productivity to be
significantly affected by macroeconomic demand conditions.
Hypothesis H1a: The productivity effect of product innovations is independent of the business cycle.
Instead, we expect the direct demand effect to vary with the business cycle. In line with the literature
presented in Section 2.1, we assume that innovators are more successful in selling new products
during positive growth periods. In this situation, incomes are increasing and budget restrictions are
less tight. In contrast, we expect product innovations to have a lower direct demand effect during
downturn and recession periods. Furthermore, the utilization of production capacities is pro-cyclical
(see e.g. Corrado and Mattey, 1997; Fagnart et al., 1999). Smolny (2002) finds that higher rates of
capacity utilization stimulate employment growth. Given these two stylized facts, we expect product
innovators to be more likely to expand their employment when facing demand increases due to new
products in upturn and boom periods as they already produce at high or full capacity. During
recessions, product innovations are accordingly expected to have less of an effect on employment
because direct demand effects tend to be weaker and firms are already struggling with excess capacity.
For these reasons, we expect a pro-cyclical demand effect of product innovation to always be positive.
Hypothesis H1b: The direct demand effect of product innovations is pro-cyclical and always
positive.
Hypotheses H1a and H1b capture the gross employment effect of product innovations. To obtain the
net employment effect of product innovations, it is further necessary to consider the indirect demand
effect. In case of a complementary relationship between the innovators’ new and old products, a
higher (lower) direct demand effect is also associated with a higher (lower) indirect demand effect.
6
Instead, if new and old products were substitutes, we would expect the demand for old products to
disproportionately decline during downturns and recessions compared to upturns and booms. This
decline may be reinforced by firms that face pressure to reduce their product range during downturn
and recession periods. Bernard et al. (2010), Broda and Weinstein (2010) and Bilbiie et al. (2012) find
product creation to be pro-cyclical, whereas product destruction and drop-out rates are countercyclical. This should rather affect old than new products. This means that at least some fraction of the
sluggish demand levels inherent to downturns and recessions can not only be attributed to tighter
budget constraints but also to the reduction of product variety. In total, combining the three
transmission mechanisms of product innovations (direct demand effect, productivity effect and
indirect demand effect) on employment growth, the net employment effect is ambiguous. Prior
empirical evidence, however, has demonstrated that the labor-creating effect tends to outweigh the
labor-destructing effect of product innovation leading to a positive net employment growth effect.
Therefore, we expect the net effect of product innovation in general to be positive and to vary with the
business cycle. If the net effect becomes negative, it is most likely to happen during a recession.
Overall, we expect the net effect of product innovations to be pro-cyclical and smallest in recession
periods.
Hypothesis H1c: The net employment effect of product innovations follows a pro-cyclical pattern,
and is assumed to be smallest in recessions.
There are two basic mechanisms underlying the employment growth effect of process innovations: (i)
a labor-destructing productivity effect and (ii) a labor-creating price effect. During downturns and
recessions, the lack of demand may discourage the introduction of new products but increases the
competition based on costs and prices (Spiegel and Stahl, 2014). In this situation, process innovations
play an important role in improving productivity, which goes at the cost of job losses (Pianta and
Lucchese, 2012). For a demand-inducing price-effect to occur, the cost reductions need to be passed
on to the product price. However, firms may be less inclined to reduce prices, as profits usually
decline during downturns and recessions. Therefore, we expect the labor-destructing productivity
effect to prevail during downturns and recessions. In contrast, process innovators may be less eager to
increase productivity during upturn and boom periods. Tether and Tajar (2008) find that the main
strength of process innovations is the flexibility and adaptability of production to market needs. This
orientation of process innovators may allow them to better exploit generic periods of high demand on
a large scale. Therefore, we expect that process innovators may be more focused on meeting the
market needs during booms than during recessions. This is less likely to be job-destructing. Even if
process innovators increase productivity during boom periods, we expect the labor-creating price
effect to occur, at least in part. Overall, we expect either stronger net job destruction or less net job
creation during downturns and recessions as compared to upturns and booms. This leads to the
following hypotheses:
7
Hypothesis H2a: The labor-destructing productivity effect of process innovations is strongest during
downturns and recessions, followed by upturns and booms.
Hypothesis H2b: Process innovations induce either stronger net job destruction or less net job
creation in downturns and recessions than in upturns and booms.
The literature presented in Section 2.2 suggests that the employment effect of organizational
innovations is driven by the same two mechanisms observed for process innovations. That is, a direct
productivity effect may reduce the innovators’ labor demand and a potentially counteracting price
effect may stimulate firms’ employment growth. Lundvall and Kristensen (1997) show that firms’
propensity to use organizational innovations as an efficiency-enhancing instrument is increasing in
competitive pressure. As discussed for process innovations, the competitive pressure is probably
highest during downturn and recession periods. This rather reduces labor demand. For these reasons,
we expect that the business cycle effects of organizational innovations on employment growth largely
correspond to the effects of process innovations.
Hypothesis H3a: The labor-destructing productivity effect of organizational innovations is strongest
during downturns and recessions, followed by upturns and booms.
Hypothesis H3b: Organizational innovations induce either stronger job destruction or less net job
creation in downturns and recessions than in upturns and booms.
3. Empirical model
In order to test our hypotheses, we adopt the approach developed by Harrison et al. (2008, 2014) that
establishes a theoretical link between firm-level employment growth and different types of
innovations. The main virtue of the model is that it relies on innovation output indicators. This means
that it incorporates the demand situation of the respective firms, which is an important element of
firms’ labor demand. Its empirical implementation is targeted at using information provided by CIS
data. In its original form, the model has been used to analyze employment effects of product and
process innovation for European, Latin American and Chinese firms (see e.g. Benavente and
Lauterbach, 2007; Hall et al., 2008; Mairesse et al., 2011; Crespi and Tacsir, 2013; Harrison et al.,
2014). We follow Peters et al. (2013) and Damijan et al. (2014) and extend the model by including
organizational innovations as well. Furthermore, we estimate the model for a large set of European
firms observed during a reasonably long period that covers different business cycle periods. In the
following, we briefly describe the model; for more details see Harrison et al. (2008, 2014).
The model is based on a two-product framework, i.e. a firm can produce two different (sets of)
products, at two points in time 𝑡 = (1, 2). At the beginning, at 𝑡 = 1, a firm produces a certain product
or product portfolio, which by definition is labelled as old (or existing) products. Between 𝑡 = 1 and
8
𝑡 = 2 , a firm may introduce one or more new or significantly improved products (product
innovations). The new products can (partially or totally) replace the old ones in case they are
substitutes. They enhance the demand of the old product in case of a complementary relationship.
Hence, at the end of that intermediate period, 𝑡 = 2, the firm produces either only old products, only
new products or both types of products.
To produce the respective output, we assume an underlying production function that is linear
homogeneous in the conventional inputs labor, capital and material. In addition, the final output
depends on Hicks neutral productivity of the respective product 𝑗 at time 𝑡, captured by 𝜃𝑗𝑗 . With
respect to old products, a firm can increase the production efficiency between 𝑡 = 1 and 𝑡 = 2 by
implementing process and organizational innovations. In addition to a firm’s own innovation-related
productivity improvements, productivity gains may be caused by learning effects, spillovers, inputs of
higher quality, training, selling or shutting down of unprofitable business units or mergers and
acquisitions. As by definition new products are not produced at 𝑡 = 1, firms cannot improve the
productivity of new products. However, it is important whether or not the productivity of new
products will be higher or lower compared to the one of old products. Based on these considerations,
Harrison et al. (2008, 2014) derive the following labor demand equation (for ease of presentation, firm
indices 𝑖 and time indices 𝑡 are suppressed:
(1)
𝑙 = 𝛼 + 𝑦1 + 𝛽𝑦2 + 𝑢
Employment growth 𝑙 originates from four main sources in the model: (i) efficiency gains in the
production of old products, 𝛼, (ii) the growth rate of the real output of old products, 𝑦1 , (iii) the real
output growth rate due to new products, 𝑦2 and (iv) the relative productivity of new products, 𝛽. 5 The
error term 𝑢 captures unanticipated productivity shocks in the production of old products at 𝑡 = 2. 6
The output growth of old products, 𝑦1 , is likely to depend on the demand for new products, at least to
some degree. That is, it captures indirect demand effects: Negative output growth will arise if new
products are substitutes to old products (cannibalization effect), whereas the growth rate will be
stimulated in case of a complementary relationship. Furthermore, the effect of 𝑦1 also captures (i)
demand changes provoked by innovations introduced by competitors (business stealing effects), (ii)
demand increases due to innovation-related price reductions (price effect), (iii) changes in consumer
preferences, (iv) policy-induced demand changes and (v) business cycle effects. Data limitations
Please note that as new products have not been produced at 𝑡 = 1, 𝑦2 cannot measure the real output growth of new
products. Instead, 𝑦2 measures the output of the new products (excluding unanticipated shocks) at 𝑡 = 2 relative to the output
of the old products at 𝑡 = 1. Therefore, it captures the real output growth rate due to new products.
5
The production functions of old and new products at 𝑡 = 2 include unanticipated productivity shocks 𝑢 and 𝜐, respectively.
In deriving equation (1), Harrison et al. (2014) show that 𝜐 cancels out. The production functions of both products also allow
for unobserved firm fixed effects, 𝜂𝑖 in every period. These firm fixed effects also vanish in the growth rate formulation.
6
9
restrict us in disentangling the underlying effects from each other. However, the data we use allow us
at least to separate the business cycle effects from the other demand-side effects.
The term 𝛽𝑦2 captures the gross employment growth induced by new products, which consists of two
channels. The first one refers to the demand growth due to new products (relative to the old products),
𝑦2 . The second one, 𝛽, involves the productivity effect of new products. This relative productivity is
defined as 𝛽 = 𝜃11 /𝜃22, i.e. as efficiency of old products in 𝑡 = 1, 𝜃11, relative to the efficiency of
new products in 𝑡 = 2, 𝜃22. Ceteris paribus, new products will generate higher employment growth if
their production is less efficient than the production technology of the old products, i.e. if 𝜃22 < 𝜃11.
In contrast, new products will induce relatively less labor demand if new products are produced more
efficiently, that is for 𝛽 < 1. According to our hypothesis H1a, we expect 𝛽 to be independent of the
business cycle.
In principle, an increase in the efficiency in the production of old products reduces firms’ labor
demand. Hence, we expect 𝛼 to be negative. Harrison et al. (2014) suggest separating the non-
innovation-related and innovation-related efficiency improvements. With respect to the latter, they
only account for process innovation-induced improvements. We extend the model by separately
investigating the employment impact of efficiency improvements of organizational innovations as
well. Rewriting equation (1) yields:
(2)
𝑙 = 𝛼0 + 𝛼1 𝑝𝑝 + 𝛼2 𝑜𝑜𝑜𝑜 + 𝑦1 + 𝛽𝑦2 + 𝑢
In addition to equation (1), equation (2) disentangles the productivity effect of old products into three
components: 𝛼0 , 𝛼1 and 𝛼2 . The first effect, 𝛼0 , represents the average efficiency gains of old
products not related to innovations. The components 𝛼1 and 𝛼2 measure the productivity effect, i.e.
gross effect, of process (𝑝𝑝) and organizational (𝑜𝑜𝑜𝑜) innovations, respectively.
Unfortunately, we cannot estimate equation (2) as we cannot observe real output growth rates in our
data. Instead, we replace the unobserved real growth rates by the observable nominal growth rates
measured as sales growth. This yields the following equation:
(3)
𝑙 − (𝑔1 − 𝜋�1 ) = 𝛼0 + 𝛼1 𝑝𝑝 + 𝛼2 𝑜𝑜𝑜𝑜 + 𝛽𝑔2 + 𝜀
The nominal sales growth of the old products, 𝑔1 , and of the new products, 𝑔2 , are defined as 𝑔1 =
𝑦1 + 𝜋1 and 𝑔2 = (1 + 𝜋2 )𝑦2 . The coefficient of the real output growth of old products, 𝑦1 , is equal
to one and can be subtracted from 𝑙. The variables 𝑔1 and 𝑔2 can be calculated by using CIS data
presented in Section 4.1. The sales growth rate of the old products, 𝑔1 , is defined as the total sales
growth rate minus the sales growth rate due to new products. The term 𝜋1 measures the unobserved
price growth rate of old products at the firm level. Potential data sources usually do not provide price
10
data on a firm level. Therefore, we proxy 𝜋1 by the price growth rate of old products at a 2-digit
industry level, 𝜋�1 . The firm-level indicator 𝜋2 is defined as the price difference between new products
at 𝑡 = 2 and old products at 𝑡 = 1 in relation to the price of the old products at 𝑡 = 1. The problem is
that we cannot observe this price information, not even on an industry level. However, substituting a
real by a nominal growth rate requires price growth information to adequately estimate the effect. As a
result, our estimation of 𝛽 suffers from an endogeneity bias caused by measurement errors and we use
an instrumental variable (IV) approach to deal with this endogeneity problem. 7 A similar problem
would arise for the estimated coefficients of 𝛼1 and 𝛼2 if 𝜋�1 was a weak proxy variable for 𝜋1 . 8 In this
case the price growth on an industry level would substantially diverge from the firm-level price
growth. Therefore, the new error term incorporates these potential sources of endogeneity, so that
𝜀 = −𝐸(𝜋1 − 𝜋�1 ) − 𝛽𝜋2 𝑦2 + 𝑢.
Even in the case 𝜋�1 is a good proxy variable for 𝜋1 , what we assume, the estimates of our innovation
indicators can still suffer from an endogeneity bias. For instance, innovations are typically the result of
investment decisions. If those decisions were correlated with any unobserved productivity shocks 𝑢
appearing at 𝑡 = 2 the estimated coefficients would indeed be biased. However, those decisions
usually take place before the realization of the shocks, i.e. before 𝑡 = 2. This means that we do not
expect serious endogeneity problems due to simultaneity. This has been confirmed by Harrison et al.
(2014) who tested the exogeneity assumption of process innovations for manufacturing firms of four
different European countries and did not find evidence for process innovations to be an endogenous
explanatory variable in the econometric model.
Overall, subtracting the proxy for the real output growth of old products, (𝑔1 − 𝜋�1 ), from employment
growth, 𝑙, allows us to estimate the gross effect of process, organizational and product innovations.9
Indeed, we cannot directly estimate the indirect demand effect, thus the net employment growth effect
(H1c) of product innovations. In order to do so, we would need more detailed demand data to
disentangle the different components of changes in 𝑦1 . However, as we will explain in more detail in
Section 4.2 and Section 4.4, we will use a decomposition analysis to show the indirect and net
employment growth effect for product innovators. Unfortunately, this is not possible for the net effect
of process and organizational innovations. Therefore, the analysis of hypotheses H2b and H3b remain
open for future research.
7
We explain our identification and estimation strategy below in Section 4.3.
The estimates are only unbiased if 𝜋
�1 corresponds to 𝜋1 . We would underestimate the productivity effects of process and
organizational innovation, otherwise.
8
Instead of using 𝑙 − (𝑔1 − 𝜋�1 ) as dependent variable, we could have used 𝑙 as dependent variable and (𝑔1 − 𝜋�1 ) as
additional explanatory variable. However, in line with the model, we would have restricted the coefficient to be one, which
would have generated the same results. Therefore, we still can interpret the results in terms of employment growth.
9
11
4. Data and estimation method
4.1. Data
We use micro data from the European-wide Community Innovation Survey (CIS). 10 This survey rests
on a European-wide harmonized questionnaire. It is biannually conducted by the national statistical
offices or legalized national institutions of the European Union’s member states, Iceland and
Norway. 11 The CIS applies the definitions and methodology of the Oslo Manual on innovation surveys
(see OECD and Eurostat, 2005). The target population covers all legally independent enterprises with
10 or more employees in manufacturing, mining, energy and water supply and selected services. The
survey collects data on firms’ innovation expenditures, different innovation output indicators and other
business-related information, e.g. employment and sales. Each CIS wave covers a three-year period.
Hence, we have calculated all our growth rates between t and t-2.
We employ five waves of CIS data that cover the years 1998-2000 (CIS3), 2002-2004 (CIS4), 20042006 (CIS2006), 2006-2008 (CIS2008) and 2008-2010 (CIS2010). For the empirical analysis, we
focus on employment effects of innovation in manufacturing firms only. Furthermore, we define two
samples. Sample 1 includes 201,691 firm-year observations and depicts our main sample. It is used for
estimating the econometric model with organizational innovation (equation (3)). One limitation of the
data is that only a limited number of countries provided information on organizational innovation in
CIS2006 as this question was not compulsory in this wave (see Table A-1 in the Appendix). As the
period 2004-2006 marked a boom period in many European countries, the drop in the number of
observations is particularly strong for the boom sample. To check whether this substantially affects the
estimated employment effects of product and process innovation, we additionally estimate the model
excluding organizational innovation. This allows us to use sample 2 that contains 225,544 firm-year
observations. Table 1 gives an overview of the distribution of the CIS waves. The distribution among
the CIS waves shows that the first three CIS waves exhibit the smallest sample sizes, whereas more
than half of the observations stem from CIS2008 and CIS2010 in sample 1.
Within the manufacturing sector, the manufacturing of basic and fabricated metals, food and
beverages as well as the textile industry hold the highest shares of observations (see Table A-2 in the
Appendix). The vehicle industry along with the industries of chemicals, rubber and plastics as well as
non-metallic mineral products, have the lowest shares in our sample.
10
We accessed CIS micro data at Eurostat’s Safe Center in Luxembourg.
11
Prior to CIS2006 (2004-2006), the survey was conducted every fourth year.
12
Table 1: Distribution of CIS sample by waves
CIS-waves
Observation period
CIS 3
CIS 4
CIS2006
CIS2008
CIS2010
Total
1998-2000
2002-2004
2004-2006
2006-2008
2008-2010
1998-2010
Sample 1
N
%
40,044 19.85
43,397 21.52
13,116
6.50
52,008 25.79
53,126 26.34
201,691 100
Sample 2
N
%
40,044 17.75
43,431 19.26
35,970 15.95
52,870 23.44
53,229 23.60
225,544 100
Source: CIS3, CIS4, CIS2006, CIS2008 and CIS2010, Eurostat; own calculation.
Note: Sample 1 refers to the model that includes organizational innovation; sample 2 refers to the model that excludes
organizational innovation.
Although all EU countries are required by law12 to conduct the CIS, they are not obliged to transfer
micro data to Eurostat. In total, micro data are available for all five waves for 11 out of 26 countries.
Five countries provide micro data only for one or two waves. The sample sizes between the countries
differ substantially (see Table A-1 in the Appendix), partly due to the country size and partly because
the CIS is compulsory for the firms in some countries like France and Italy, whereas it is voluntarily in
other countries. To get representative results, we apply weighting factors to all descriptive statistics
and estimations. 13
Another limitation of the data is that firm-level observations cannot be linked between subsequent CIS
waves due to missing firm identifiers. Hence, we can only apply pooled OLS and IV estimators. In
order to account for the fact that some firms are repeatedly observed and the i.i.d. assumption may be
violated, we will use clustered standard errors at the industry-country level to allow for correlation
among the error terms within the cluster.
4.2. Variables
In accordance with the underlying model, the dependent variable, EMP, is defined as 𝑙 − (𝑔1 − 𝜋�1 ).
The employment growth, 𝑙, is measured as the relative change in the number of employees (head
counts) between and t and t-2. The real output growth due to old products, 𝑔1 − 𝜋�1 , denotes the
difference between (i) the nominal sales growth rate of old products (𝑔1 / SGR_OLDPD) and (ii) the
growth rate of prices for old products at the industry level (𝜋�1 / PRICEGR). 14 The nominal sales
growth rate due to old products (𝑔1 / SGR_OLDPD) is calculated as the total sales growth rate minus
the sales growth rate due to new products (𝑔2 / SGR_NEWPD; see below). To calculate 𝜋�1 we use
producer price indices at the 2-digit country-industry level as published by Eurostat.
12
Up to 2010, CIS has been collected under Commission Regulation (EC) No 1450/2004. From 2012 onwards, Commission
Regulation No 995/2012 applies.
13
Weighting is implemented by using sample weights that extrapolate to the population number of firms in each stratum.
14
Lower-case letters denote the model variables whereas upper-case letters describe their empirical counterparts in the data.
Remember that our growth rates are defined as the growth between t and t-2.
13
Process innovation (𝑝𝑝) is represented by the dummy variable PC that takes on the value one if firms
have introduced only process innovations but no product innovations between t and t-2. According to
the Oslo Manual, a process innovation is the implementation of a new or significantly improved
production process, distribution method, or supporting activity. This definition includes process
innovations that are introduced to support the launch of product innovations (see OECD and Eurostat,
2005, paragraph 164). According to our theoretical model, however, 𝛼1 captures efficiency gains of
process innovation related to the production of old products. Our focus on pure process innovators in
the empirical analysis allows us to identify the employment effect of process innovations related to old
products more accurately.
Organizational innovations involve new methods in the firms’ business practices, workplace
organizations or external relations. According to the Oslo Manual, they are intended to e.g. lower
administrative costs and to increase labor productivity. However, they are not considered as an enabler
for product innovations. One example of an organizational innovation is firms’ reduction of hierarchy
levels (delayering). A flatter management structure lowers costs and should increase firms’
productivity as there are fewer management hurdles to overcome within a decision process. We
measure organizational innovations (𝑜𝑜𝑜𝑜) by the indicator ORGA that takes on the value one if firms
introduced at least one organizational innovation between t and t-2.
Our key variable for measuring product innovation output is the sales growth rate due to new products
(𝑔2 ). Its empirical counterpart, SGR_NEWPD, is calculated as year t’s sales share with new products,
which have been introduced between t and t-2, multiplied by the ratio of year t’s sales divided by the
sales of t-2.
Beyond the information required by the model’s structural equation, employment growth is likely to
be influenced by a set of other characteristics. An important determinant for employment growth is
firm size. According to Gibrat’s law, firms grow proportionally and independently of firm size. In
contrast, Mansfield (1962) finds that smaller firms’ growth rates are higher and also more volatile.
Jovanovic (1982) provides a theoretical background for Mansfield’s related analyses. His model
suggests that surviving young and small firms grow faster than older and larger ones because of
managerial efficiency and learning by doing. To control for size effects we include the dummy
variables, MEDIUM – taking on the value one for firms with 50-249 employees – and LARGE, for
firms with at least 250 employees.15 Firms with less than 50 employees, SMALL, build our reference
category. All size dummies are related to the period t-2. Furthermore, we control for ownership effects
as employment grows slower and is also more volatile in foreign-owned firms (Dachs and Peters
2014). We include two dummy variables indicating that a firm belongs to a firm group that has a
domestic (DGP) and foreign headquarter (FGP), respectively. Domestic unaffiliated firms serve as
15
To further scrutinize size heterogeneity, we split our sample into SMEs and large firms (see Section 5.2).
14
reference group (DUF). 16 A set of time, industry and country dummies based on the information
presented in Table A-1 and Table A-2 is also included.
To properly examine business cycle effects, we split our estimation sample into the four phases of the
business cycle. In general, the business cycle describes fluctuations in economic activity that an
economy experiences over a period of time. A business cycle consists of four phases: upturn, boom,
downturn and recession. Our analysis uses real GDP growth rates on a country level, provided by
Eurostat. Based on that, we define our business cycle indicator as follows:
Table 2: Definition of the four business cycle phases
Business cycle phase
GDP growth is…
Upturn
increasing and positive
Boom
increasing and positive and subsequently decreasing
Downturn
decreasing but (still) positive
Recession
negative
One potential issue is the time period used to calculate this indicator. Statistical offices often use
quarterly data on GDP growth to define a business cycle. In empirical work, it is also common to
employ one-year growth rates. The CIS data covers a three-year period, in CIS2010 for instance the
period 2008-2010. Hence, we use two-year GDP growth rates, i.e. in the example above the growth
rate between 2008 and 2010.
Splitting the sample according to the business cycle phases ignores the information about the strength
of GDP growth, which varies considerably across European countries. For this reason, we include in
our estimations the information about country-level real GDP growth rates (GDPGR) between t and t2. This captures general demand effects. Firm-specific demand effects are already covered by 𝑔1 and
𝑔2 . Therefore, our equation to be estimated is the following:
(4)
𝐸𝐸𝐸𝑡|𝑡−2 = 𝛼0 + 𝛼1 𝑃𝑃𝑡|𝑡−2 + 𝛼2 𝑂𝑂𝑂𝑂𝑡|𝑡−2 + 𝛽𝑆𝑆𝑆_𝑁𝑁𝑁𝑁𝑁𝑡|𝑡−2 + 𝑋𝑡 𝛾 + 𝑣𝑡
The matrix 𝑋𝑡 includes our control variables MEDIUM, LARGE, DGP, FGP and GDPGR as well as
the time, industry and country dummies for each firm i at time t. The remaining variables, the error
term 𝜀𝑡 and 𝛾 in equation (4) denote vectors.
4.3. Descriptive statistics
About half of the firms (49.5%) can be classified as innovators having introduced at least one product
(27%), process (11%) or organizational innovation (32%) (see Table A-3 in the Appendix). This
16
See Table A-3 for a distribution of the respective firm groups included in our sample.
15
section presents basic descriptive results on the relationship between innovation and employment
growth in different phases of the business cycle. For the total sample, Figure 1 shows that the average
employment growth rates across all firms follow a pro-cyclical pattern. During upturn periods, firmlevel employment has grown by 5.9%. It has increased by 8.2% during booms, whereas the growth
rate diminished to 3.4% during downturn periods. The level of employment has been reduced by 4.4% during recessions. 17
Figure 1: Employment growth by innovation status in different business cycle periods, in %
12.0
10.0
9.2
8.2
6.0
4.0
5.9
7.9
7.5
7.1
8.0
5.8
5.5
9.6
9.3
9.5
7.3
6.6
6.2
4.8
4.8
3.4
1.8
2.0
0.0
-2.0
-2.3
-4.0
-6.0
-4.4
-3.1
-3.2
-3.2
-5.6
-8.0
Total
Innovator
Upturn
Product
Boom
Process
Downturn
Orga
Non-innovator
Recesssion
Source: CIS3, CIS4, CIS2006, CIS2008, CIS2010, Eurostat; authors’ own calculation.
Note: Depicted are weighted average two-year employment growth rates; the group “Innovator” refers to firms that have
implemented at least one process, product or organizational innovation between t and t-2.
This pro-cyclical relationship holds for each type of innovator and non-innovator, respectively. It
means that each group has suffered from employment reductions during recessions and has achieved
employment growth during the other periods. More importantly, we observe substantial differences in
average employment growth between the group of innovators and the group of non-innovators in each
phase of the business cycle. For instance, during booms, innovators have raised their employment
level by 9.2%, whereas non-innovators’ employment has increased by 6.6%. Such a strong difference
is observed for all types of innovators. The most notable difference concerns the recession period.
Innovators have decreased their employment level by -3.1% during recessions, whereas noninnovators have cut employment by almost twice as much (-5.6%). This implies that innovators,
particularly product innovators, have been more resilient to periods of negative economic growth. The
17
The employment growth rates are not directly comparable to official employment statistics. First, CIS applies the lower
threshold of 10 employees. Second, employment changes due to firms exiting and entering are not captured by CIS data
because survey response is conditional on surviving and employment growth rates are not defined for newly established
firms. Third, official statistics are based on a different calculation method which is the ratio of the sum of changes in
employment for all firms to the sum of employed personnel.
16
difference in the employment growth rates between innovators and non-innovators may be due to the
innovators’ superior adaptability to shocks (Meghir et al., 1996). Accordingly, innovating firms are
more flexible and have lower adjustment costs of employment when faced with negative shocks. Of
course, it might also be that innovators and non-innovators differ in other firm characteristics that
contribute to the better employment performance of innovators in general and to the stronger resilience
of innovators in recessions in particular. The econometric analysis is aimed at disentangling the role of
innovation while controlling for these other firm characteristics.
Figure 2 brings two key variables of the empirical model into focus: the average nominal sales growth
rates due to new and to old products. New product sales as well as old product sales follow a procyclical pattern. New product sales have increased by 8.8% during upturn periods and by 12.1%
during booms. The strong demand has slowed down to 9.3% during downturn periods. The weakest
demand has occurred in recessions, with a growth rate of 8.6%. Pairwise mean difference tests show
that the average sales growth rates due to new products are significantly different at the 1% level
between upturn and boom, boom and downturn and downturn and recession, respectively. Hence, we
find support for our hypothesis H1b. New product sales are pro-cyclical and always positive. Despite
significant differences across business cycle periods – except for the comparison between upturn and
recession – it is remarkable how much sales growth is generated with new products even in recessions.
If we focus only on product innovators, the sales growth due to new products is about 30% in the
recession, only little less than in the boom period (33%), see Table A-5. This finding will become
important in explaining employment growth differences of innovators and non-innovators in the
recession.
Sales growth due to old products is substantially lower than due to new products in all phases of the
business cycle. While the gap remains rather stable for upturn, boom and downturn – ranging from 4.5
to 7.7 percentage points – it stands out in the recession. Sales with old products have severely suffered
from a drop in demand and declined by -19%. Table A-5 in the Appendix shows that this substantial
decline is partly due to a reduction in demand for old products of product innovators (-37%). It may
have been caused by significant product cannibalization or a reduction in the innovators’ product
range. Another explanation relates to tighter budget constraints during recessions. The potential for
budget restrictions is indicated by the sales loss of non-innovators, i.e. firms that only offered old
products and did not introduce any innovation between t and t-2. Their sales growth rate decreased by
-12.5% during recessions. This decline is substantially larger than the net loss in sales of product
innovators (-7.1%).
17
Figure 2: Sales growth due to new and old products, in %
15.0
12.1
10.0
5.0
9.3
8.8
2.4
4.4
8.6
4.8
0.0
-5.0
-10.0
-15.0
-20.0
-19.0
-25.0
Upturn
Boom
Sales growth - old products
Downturn
Recesssion
Sales growth - new products
Source: CIS3, CIS4, CIS2006, CIS2008, CIS2010, Eurostat; own calculation.
Note: Weighted figures.
4.4. Estimation approach and identification
As discussed in Section 3, the estimation of the relative productivity effect of product innovation, 𝛽, is
subject to a measurement error of the sales growth rate due to new products. Therefore, we employ a
weighted instrumental variable (IV) approach to estimate equation (4) consistently. In our case,
variables qualifying as instruments should be correlated with the sales growth due to new products
(i.e. innovation success) and should be uncorrelated with the error term. In particular, the instruments
have to be uncorrelated with the relative price difference of new and old products. We cannot use any
lagged values of the endogenous right-hand variable as instrument because Eurostat does not provide
firm identifiers. Instead, our identification strategy is based on three external instruments. We expect
these variables to be important in explaining innovation success and to be uncorrelated with the
relative price difference of new to old products.
Following Harrison et al. (2014), we use RANGE as instrument. RANGE is a binary indicator
measuring whether the implemented product innovations have been aimed at increasing the product
range or not. We assume RANGE to be correlated with the expectations of new product sales.
Enlarging the range of products is a strategic decision that may require more resources than other
product innovations. We expect that those firms put more effort into the development and the market
introduction. This means, they may spend more on e.g. R&D, the product design and marketing
activities. If firms invest more for product innovations it should also have a positive effect on the new
products’ success. Enlarging the range of products, however, does not imply any particular direction
of the changes in prices. New products added to the firm’s product portfolio might be of higher
(lower) quality sold at higher (lower) prices than existing products of similar quality and price. Our
18
second instrument is a binary indicator as well. It takes on the value one if the firm actively cooperated
in innovation projects with other agents, COOP. Other agents include, for instance, suppliers, research
institutions and competitors. Firms benefit from knowledge spillovers in cooperation projects. We
expect these spillovers to increase the likelihood of successfully developing new products which
should be in turn correlated with higher expected new product sales. Furthermore, cooperating firms
may establish new ways (or channels) of distributing the products. This should also be correlated with
higher expected sales with new products. However, we do not expect this to significantly affect the
relative price between old and new products. Our third instrument, RD, indicates whether a firm
performs R&D activities on a continuous base or not. A continuous research effort should be related to
a higher likelihood of inventing new technologies and as a result of higher expected sales with new
products. However, we do not expect a more continuous research effort to be substantially linked to
the relative price difference.
5. Empirical evidence on employment effects of innovation over
the business cycle
5.1. Full sample results
5.1.1.
Regression results
Table 3 presents the regression results of the basic OLS estimations. We split the sample into the four
phases of a business cycle, i.e. upturn, boom, downturn and recession. The coefficient of the sales
growth rate due to new products (SGR_NEWPD) is central to our analysis on the relationship between
employment growth and product innovation.
The coefficient measures the average employment effect of the relative efficiency of the production
between old and new products. The results show that the relative production efficiency, 𝛽̂, is smaller
than one. This means, producing new products is more efficient than producing old products resulting
in less labor demand. Efficiency gains range between 9.7% and 13.4%. To test whether these estimates
are significantly different from one, we performed Wald tests. Each displayed p-value of the Wald test
suggests that the relative production efficiency is significantly smaller than one. However, we suspect
SGR_NEWPD to be endogenous due to measurement error and hence 𝛽̂ to be downward biased. To
address endogeneity, we conduct IV estimations as explained in Section 4.4.
19
Table 3: Employment effects of innovation over the business cycle, OLS estimations
Business cycle phase
Downturn
Dep. var.: EMP
Upturn
Boom
SGR_NEWPD
Observations
R2_adjusted
0.866***
(0.014)
-3.090***
(0.817)
-0.967**
(0.398)
3.727***
(0.554)
-2.906***
(0.470)
-4.263***
(0.613)
-0.957
(0.708)
-0.787
(0.791)
-67.655***
(7.270)
67,468
0.383
0.892***
(0.015)
-1.374
(1.310)
1.360*
(0.693)
-0.509**
(0.202)
0.084
(0.859)
-3.135**
(1.216)
3.330***
(1.164)
0.969
(1.159)
1.383
(1.729)
15,863
0.495
0.879***
(0.011)
-3.486***
(0.846)
0.376
(0.548)
-0.611***
(0.177)
-0.845
(0.556)
-0.777
(0.725)
0.629
(0.655)
0.237
(0.657)
-14.920***
(2.614)
67,179
0.393
0.903***
(0.013)
-1.188
(0.887)
0.187
(0.457)
-0.362*
(0.217)
-1.910***
(0.499)
-3.792***
(0.648)
1.327*
(0.676)
-1.730***
(0.650)
2.972*
(1.600)
51,181
0.467
Further test
Wald-test for β=1 (p-value)
0.000***
0.000***
0.000***
0.000***
PC
ORGA
GDPGR
MEDIUM
LARGE
DGP
FGP
Constant
Recession
Note: * p<0.1; ** p<0.05; *** p<0.01; sample 1; weighted OLS estimations; clustered standard errors in parentheses
(clustered by 2-digit industry and country); time, industry and country dummies are included and each set of dummies is
jointly significant.
Our main results, the IV estimates, are presented in Table 4. The test on exogeneity corroborates our
suspicion that the sales growth due to new products is endogenous. The tests reject the null hypothesis
of exogeneity for all business cycle phases at least at the 5% level. The IV results disclose positive and
significant estimates for 𝛽̂ across all phases of the business cycle. These estimates are larger than the
OLS estimates. Except for the downturn period, the coefficients are only slightly smaller than one. The
Wald tests confirm that each of the four coefficients is not significantly different from one. Hence,
new and old products are produced equally efficient, on average. According to the structural model,
the coefficient indicates that a 1% increase in the sales due to new products leads to a 1% increase in
gross employment, independent of the cyclical situation. The finding that 𝛽̂ is not significantly
different from one in all four estimates confirms our hypothesis H1a stating that the productivity effect
of product innovation is independent of the business cycle.
20
Table 4: Employment effects of innovation over the business cycle, IV estimations
Business cycle phase
Downturn
Dep. var.: EMP
Upturn
Boom
SGR_NEWPD
0.991***
(0.022)
-1.665**
(0.821)
-2.284***
(0.463)
3.637***
(0.555)
-3.090***
(0.463)
-4.728***
(0.612)
-1.487*
(0.792)
-1.113
(0.804)
-67.158***
(7.291)
67,468
0.377
0.971***
(0.031)
-0.173
(1.401)
0.536
(0.743)
2.823
(1.812)
-0.014
(0.863)
-3.577***
(1.284)
3.203***
(1.165)
1.039
(1.147)
-33.436**
(15.817)
15,863
0.492
1.003***
(0.026)
-1.816*
(0.941)
-1.393**
(0.614)
-0.600***
(0.175)
-1.260**
(0.597)
-1.358*
(0.791)
0.566
(0.648)
0.123
(0.659)
-15.094***
(2.647)
67,179
0.387
0.988***
(0.029)
-0.214
(1.038)
-0.716
(0.501)
-0.017
(0.278)
-2.026***
(0.503)
-3.976***
(0.667)
1.244*
(0.663)
-1.825***
(0.628)
3.013*
(1.647)
51,181
0.464
0.691
0.349
0.908
0.681
0.000***
0.004***
0.000***
0.024**
0.375
0.311
0.160
0.220
0.237
0.469
0.162
0.132
0.340
0.413
0.264
0.056*
0.809
0.163
0.929
0.112
21.885***
(0.771)
4.984***
(0.801)
10.112***
(1.279)
436.24***
20.747***
(1.226)
4.852***
(1.808)
7.172***
(1.298)
228.91***
23.241***
(0.956)
5.602***
(0.770)
5.731***
(1.392)
321.22***
19.504***
(1.027)
2.717**
(1.104)
5.937***
(1.439)
265.71***
297.655***
61.597***
1326.516***
706.997***
560.833***
300.275***
801.494***
346.558***
926.194***
65.384***
462.643***
45.523***
986.760***
87.832***
335.099***
48.947***
PC
ORGA
GDPGR
MEDIUM
LARGE
DGP
FGP
Constant
Observations
R2_adjusted
Further tests
Wald-test for β=1 (p-value)
Test on exogeneity (p-value)
SGR_NEWPD
Test on instrument validity
Hansen J test (p-value)
Difference-in-Hansen test (p-value)
RANGE
COOP
RD
First stage results of SGR_NEWPD
RANGE
COOP
RD
F test on excluded instruments (F)
Test on underidentification
Kleibergen-Paap LM test (chi2)
Test on weak instruments
Kleibergen-Paap F test (F)
Inference robust to weak identification
Anderson-Rubin Wald test (chi2)
Stock-Wright LM test (chi2)
Recession
Note: * p<0.1; ** p<0.05; *** p<0.01; sample 1; weighted IV regressions; clustered standard errors in parentheses; time,
industry and country dummies are included and each set of dummies is jointly significant.
The coefficient of PC measures the productivity effect (gross employment effect) of process
innovations. The significantly negative coefficients in the upturn and downturn period indicate that the
implemented process innovations have been labor-destructing, on average. The coefficients are
negative, though insignificant for booms and recessions. The finding for the boom period is in line
21
with our hypothesis H2a. We expect either no or at least the smallest labor-destructing productivity
effect during booms because process innovations are more likely to be aimed at increasing capacity
and flexibility than efficiency in this situation. The insignificant estimate for the recession, however, is
counterintuitive as we expected stronger negative effects for downturns and recessions. For this
reason, we can only partially confirm hypothesis H2a. The employment effects of organizational
innovations are very similar to the ones of process innovations. That is, as expected organizational
innovations significantly reduce the innovators’ labor demand during upturn and downturn periods.
The positive though insignificant effect in the boom period is again in line with hypothesis H3a. Like
for process innovation, we do not find significant labor-displacing effects of organizational
innovations in the recession. This similarity may be attributable to the assumption that organizational
and process innovations rely on the same channels to influence employment growth. As a result, we
only partially confirm hypothesis H3a as well.
With respect to the control variables, an interesting finding is that affiliates of foreign multinational
firms (FGP) grow with a significantly lower rate reduced during recessions than domestically owned
firms. An explanation could be that foreign-owned firms are more exposed to fluctuations of the world
market via exporting activities. In addition, the negative coefficient may also imply that multinational
firms rather prefer to lay off employees abroad than at home during recessions. The results
furthermore corroborate that firm size matters for employment growth. In all business cycle phases,
large firms (LARGE) and medium-sized firms (MEDIUM) have lower employment growth rates than
the reference category of small firms, even though the effect of MEDIUM for booms is not significant.
This is in line with findings of Mansfield (1962) and the predictions of Jovanovic (1982).
Furthermore, higher GDP growth is associated with significantly higher employment growth rates
during upturn periods. The negative effect of GDP growth for downturn periods is, however, a bit
puzzling. One explanation could be that a relatively high GDP growth rate within a downturn period
may mean that the downturn “has just” started. 18 In this situation, firms may anticipate a downturn
period. As a result, they may become more reluctant in hiring new employees at the onset of an
economic downturn, which would reduce the firms’ employment growth. This partly prevents the
situation of dismissing employees due to excessively high labor costs if product demand further
decreases. The national labor markets of EU countries are relatively well protected against dismissals.
For instance, many of the EU countries, which are also members of the OECD, have strict
employment protection laws (see OECD, 2013, Chapter 2). Hence, it is not easily feasible for firms to
lay off employees. The insignificant effect for recession periods implies that once GDP growth
becomes negative the exact level of the economic slump does not significantly affect the firms’ labor
demand.
18
Remember that according to our definition of a downturn, GDP growth is decreasing but not negative (see Table 2).
22
We carefully test our identification strategy using various tests. The consistency of the IV estimator
depends on the validity of instruments. Therefore, we perform a Hansen J test on overidentifying
restrictions for overall instrument validity. As we use three instruments, we also perform the
Difference-in-Hansen C test to test for exogeneity of a single instrument. Using a conventional level of
significance of 5%, all tests confirm that our instrument set is valid (see Table 4, Table 5 and in
section 5.2.1 Table 6 as well). 19 In addition, we check for non-weakness of the instruments. Weak
instruments can lead to a large relative finite-sample bias of IV compared to the bias of OLS. All first
stage regressions show our instruments to be strongly correlated with SGR_NEWPD, as is also
supported by the significant test statistics of the Anderson-Rubin Wald test and the Stock-Wright LM
test. Furthermore, the F test of excluded instruments always yields a statistic clearly being larger than
ten. The regression output tables also display the Kleibergen-Paap LM test on underidentification as
well as the F test proposed by Kleibergen and Paap (2006). All these tests indicate our instruments to
be neither invalid nor weak.
As explained in Section 4.1, the information on organizational innovation is only available for very
few countries in CIS2006. Including this indicator in our estimations reduces our sample by about
24,000 observations. This affects primarily the boom period. For this reason, we exclude
organizational innovation and re-estimate our model using the larger sample 2 to check for substantial
differences in the estimates. Table 5 presents the results of these regressions. Overall, the estimates are
very robust. The main difference compared to Table 4 is that the estimated employment growth effect
of SGR_NEWPD slightly weakens. However, based on the corresponding Wald test we still find all of
them not to be significantly different from one on the conventional 5% level. Hence, the overall
conclusion of hypothesis H1a remains, i.e. the productivity effect of product innovations is
independent of the business cycle.
19
At the 10% level, overall instrument validity cannot be rejected either. However, COOP and RANGE are only valid on a
5% level for downturns in Table 4, Table 5 and Table 6 and in booms in Table 5, respectively.
23
Table 5: Employment effects of innovation over the business cycle (excluding organizational
innovation), IV estimations
Business cycle phase
Downturn
Dep. var.: EMP
Upturn
Boom
SGR_NEWPD
0.966***
(0.020)
-2.283***
(0.799)
3.703***
(0.551)
-3.113***
(0.450)
-4.906***
(0.587)
-1.908**
(0.786)
-1.430*
(0.779)
-64.522***
(7.871)
70,396
0.381
0.960***
(0.023)
-0.698
(0.935)
-1.179***
(0.456)
-0.004
(0.714)
-2.818**
(1.106)
2.714***
(0.986)
1.208
(0.980)
3.931
(4.476)
31,345
0.463
0.980***
(0.022)
-2.080**
(0.841)
-0.072
(0.219)
-1.386**
(0.544)
-1.755**
(0.762)
0.427
(0.616)
-0.385
(0.621)
-21.702***
(3.215)
72,519
0.377
0.980***
(0.027)
-0.359
(0.988)
-0.006
(0.277)
-2.095***
(0.515)
-4.162***
(0.685)
1.185*
(0.668)
-1.881***
(0.615)
2.973*
(1.639)
51,284
0.463
0.079*
0.077*
0.357
0.455
0.000***
0.001***
0.000***
0.024**
0.319
0.245
0.153
0.223
0.258
0.323
0.139
0.095*
0.366
0.327
0.303
0.054*
0.703
0.196
0.959
0.102
22.796***
(0.779)
5.818***
(0.812)
11.241***
(1.243)
452.6***
21.910***
(1.043)
5.724***
(1.460)
8.168***
(1.156)
320.64***
24.991***
(0.943)
6.180***
(0.697)
6.320***
(1.341)
342.75***
20.380***
(1.024)
3.470***
(1.107)
6.495***
(1.463)
245.55***
347.758***
145.609***
1707.381***
843.865***
648.335***
527.052***
976.183***
423.313***
1054.433***
76.322***
667.648***
49.252***
1111.730***
103.455***
352.731***
48.449***
PC
GDPGR
MEDIUM
LARGE
DGP
FGP
Constant
Observations
R2_adjusted
Further tests
Wald-test for β=1 (p-value)
Test on exogeneity (p-value)
SGR_NEWPD
Test on instrument validity
Hansen J test (p-value)
Difference-in-Hansen test (p-value)
RANGE
COOP
RD
First stage results of SGR_NEWPD
RANGE
COOP
RD
F test on excluded instruments (F)
Test on underidentification
Kleibergen-Paap LM test (chi2)
Test on weak instruments
Kleibergen-Paap F test (F)
Inference robust to weak identification
Anderson-Rubin Wald test (chi2)
Stock-Wright LM test (chi2)
Recession
Note: * p<0.1; ** p<0.05; *** p<0.01; sample 2; weighted IV regressions; clustered standard errors in parentheses; time,
industry and country dummies are included and each set of dummies is jointly significant.
24
5.1.2.
Employment decomposition
The main specification we estimate by the IV approach allows us to identify the gross employment
effect of product, process and organizational innovation. We complement our estimation results with a
decomposition analysis. This allows us to quantify the absolute contribution of different sources to
average employment growth for different types of firms. In particular, we are able to disentangle the
employment effects of product, process and organizational innovations from effects originating from
general demand and productivity trends. We follow the decomposition procedure proposed by
Harrison et al. (2014) and Peters et al. (2013):
(5)
[1 − 𝐼(𝑔2 > 0)] ������������
(𝑔1 − 𝜋�1 ) + �������������
(𝑔1 − 𝜋�1 ) + �������������
𝑙 ̅ = 𝛼�0 + 𝛼�1 ���
𝑝𝑝 + 𝛼�2 𝑜𝑜𝑜𝑜
������� + ��������������������
𝐼(𝑔2 > 0) ������������
𝐼(𝑔2 > 0)𝛽̂ 𝑔̅2
In equation (5), bars denote mean values. Hence, the first term, 𝛼�0 , measures the contribution of the
general productivity trend in the production of old products to average employment growth, 𝑙 ̅. It
accounts for all changes in efficiency and employment that are not attributable to firms’ own
innovations. For instance, 𝛼�0 captures average employment effects of training, improvements in the
human capital endowment and productivity effects from spillovers. The general productivity trend
captures the general industry-, country-, time-, size-, GDP- and ownership-specific productivity trend.
It is measured as the average effect across innovators and non-innovators. The second (𝛼�1 ) and the
third (𝛼�2 ) terms capture the productivity effects of the share of process and organizational innovators,
respectively.
The term �������
𝐼(𝑔2 ) measures the proportion of product innovators, whereas ������������
1 − 𝐼(𝑔2 ) measures the share
��������������������������
of non-product innovators. This implies that the fourth component, [1
− 𝐼(𝑔2 )](𝑔1 − 𝜋�1 ) captures the
average employment growth caused by the average real growth of old product demand of the share of
non-product innovators. A demand increase of old products can be due to a change in consumers’
preferences, price reductions but also due to rivals’ product innovations (business stealing effect). In
������������
contrast, the fifth term �������������
𝐼(𝑔2 > 0) (𝑔
� 1 ) accounts for indirect effects related to the demand for old
1−𝜋
products of the proportion of product innovators. These indirect demand effects may reflect
������������
(𝑔1 − 𝜋�1 ) < 0 , and complementary effects, (𝑔
cannibalization effects, ������������
� 1 ) > 0 , respectively.
1−𝜋
�������������
𝐼(𝑔2 > 0)𝛽̂ 𝑔̅2 measures the average employment growth due to average increases in the demand for
new products of the share of product innovators. The sum of the fifth and sixth term denotes the net
contribution of product innovators to employment growth.
To sum up, we calculate this decomposition by inserting the estimated coefficients 𝛼�0 , 𝛼�1 , 𝛼�2 and 𝛽̂,
the weighted shares of process, organizational, non-product and product innovators and weighted
averages for employment, price and sales growth rates (either for all firms or for the corresponding
group of firms). The decomposition analysis shows that the net contribution of product innovations to
average employment growth depends on (i) the demand increase for new products, 𝑔̅2 (“innovation
25
success”), (ii) the estimate of the relative production efficiency between old and new products, 𝛽̂, (iii)
(𝑔1 − 𝜋�1 ) , and (iv) the proportion of product
possible shifts in the demand for old products, ������������
�������������
innovators 𝐼(𝑔
2 > 0) .
Figure 3 provides a graphical illustration of the decomposition of average employment growth during
the four phases of a business cycle. The sources (i) general productivity trend in the production of old
products, (ii) productivity effect of process innovations and organizational innovations, (iii) output
growth of non-product innovators due to old products and (v) the net employment contribution of
product innovations sum up to the total average employment growth, which is also presented in Figure
1. 20 The figure further splits the net contribution of product innovation into the increase in output due
to new products and shifts in demand for old products.
Figure 3: Contribution of innovation types to employment growth over the business cycle, in %
Employment growth - decomposed into:
General productivity trend in production of old products
Contribution of process and organizational innovation
Output growth due to old products
Net contribution of product innovation
thereof output reduction in old products
thereof output increase due to new products
15.0
11.7
8.7
10.0
8.2
9.3
8.4
8.5
7.3
5.9
5.6
5.3
4.3
3.4
5.0
2.6
2.7
0.2
0.0
-1.5
-0.9
-0.2
-0.6
-2.4
-2.8
-5.0
-4.4
-5.9
-5.7
-6.4
-6.7
-10.0
-9.1
-10.9
-15.0
Upturn
Boom
Downturn
Recession
Source: CIS3, CIS4, CIS2006, CIS2008, CIS2010, Eurostat; own calculation.
Note: Decomposition of the weighted-average two-year employment growth rate; the decomposition is based on the
regression results presented in Table 4.
This figure reiterates the strictly pro-cyclical pattern of the average employment growth, as was shown
in Figure 1. Most importantly, it discloses that the net contribution of product innovation (dark grey
bar) on employment growth is strictly pro-cyclical as well. This implies that the net employment effect
20
We combine the effects of process and organizational innovations because their separate contributions are very small in
size.
26
of product innovation is tightly connected to the specific macroeconomic growth period. According to
this figure, product innovation has increased average employment by 2.7% during upturns. This shows
that product innovation creates much more employment due to the demand effect than it destroys due
to the productivity effect and substitution effect between old and new products. This effect is even
stronger in an economic boom (+5.3%). In downturns, the gross employment creation effects of
product innovation (white-dotted grey bar) shrink, but the net effect remains positive and of similar
size than in upturns (2.6%). Only in the recession, the net contribution of product innovation is
negative because output from new products does not grow fast enough to compensate losses in old
products (-2.4%). These findings are in line with our hypothesis H1c. This pro-cyclicality can be
explained by the proc-cyclical direct demand effect of product innovation. But severe product
cannibalization and product range reduction also play an important role for this finding. In every
business cycle phase, the introduction of new products is associated with a decline in the demand for
old products (white-striped grey bar). This fall is rather low in periods of positive economic growth (5.9% and -6.4%) but becomes much larger in recessions (-10.9%). Overall, our findings reveal that
new products are an important driver for job creation in upturns, booms and downturns.
However, Figure 3 also reveals that – except for the recession – the main source for job creation is
output growth due to old products of non-product innovators (light grey bar). Its contribution to
average employment growth amounts to 5.6 percentage points during upturns, 8.4 during booms and
4.3 during downturns. At first glance, this finding might be counter-intuitive as the total sales growth
rates of product innovators are larger across all periods than those of non-product innovators (Table A5) and given that we have found the same efficiency between old and new products. The larger
employment contributions of old products are due to a much larger share of non-product innovators
(73%) than product innovators (27%; see Table A-3).
The second novel and intriguing finding of our analysis is that product innovators are more resilient to
recessions than non-product innovators. Product innovators are able to compensate employment
losses caused by demand reductions of old products by demand gains of new products. This ability is
particularly valuable during recession periods. While product innovators have experienced a similar
decline in employment due to the lower demand for old products (-10.9%) than non-product
innovators (-9.1%), their employment gains from additional demand for their new products (+8.5%).
This demand growth due to new products has remained remarkably high during the recession (see also
Figure 2). As a result net employment losses of product innovators are significantly lower (-2.4%) in
recessions than those of non-product innovators (-9.1%). Therefore, product innovations seem to have
an employment-preserving effect and make the innovators more resilient when facing negative
economic shocks. Labor-destructing productivity effects of process and organizational innovations do
not affect employment growth on a large scale. Taken together, both types of innovations are weakly
labor-creating during booms and they modestly displace labor demand during the three other business
cycle phases, primarily during upturns and downturns.
27
Another factor dampening employment fluctuations over the business cycle is the general productivity
trend in the production of old products. The general productivity trend curbed employment growth
during upturns, booms and downturns, but it is strikingly positive during recessions (7.3%). Hence,
results indicate firms’ tendency towards labor hoarding during recession periods. This means, they
reduce their staff by less than the demand for their products has fallen (Bhaumik, 2011). In general,
labor hoarding results in a decrease of productivity. Our finding is in line with other studies. Leitner
and Stehrer (2012) observe frequent labor hoarding during the recent crisis in Central and Eastern
European countries. Labor hoarding may have occurred or intensified by effective short-time work
programs during the 2008/2009 recession many EU governments have offered. Balleer et al. (2016)
for Germany and Boeri and Brücker (2011) for a set of European countries find that short-time work
had a positive effect in terms of job savings during the recent recession, even though it did not save as
many jobs as the utilization of short-time work could have saved (Boeri and Brücker, 2011).
5.2. Results for SMEs and large firms
5.2.1.
Regression results
Small and medium-sized enterprises have been found to be the most important driver for employment
growth (see Neumark et al., 2011; Haltiwanger et al., 2013). In particular, small enterprises are more
flexible to react to new opportunities and are able to survive in niche markets. They primarily benefit
from the personal engagement of the entrepreneur who transfers her knowledge on technologies and
markets (see e.g. Thurik, 2009 for an overview). At the same time, there are many arguments put
forward in the literature why large firms exhibit advantages in the innovation process (see e.g.
Kleinknecht, 1989; Cohen, 1995, 2010). They have more internal financial means and better access to
external funding to finance innovation projects more easily. Large and diversified firms also have
more potential application possibilities for new knowledge (Rosenberg, 1990). Data from the recent
economic crisis provides evidence that innovation activities in larger firms have been less affected by
the recession. This supports the view that large firms have advantages in the innovation process (see
e.g. Paunov, 2012; Rammer, 2012; Archibugi et al., 2013). This speaks in favor of a larger
contribution of innovation to employment growth in large firms, in particular in the recession. To
directly account for size differences in the employment effects of innovations, we split our sample into
two size classes, i.e. SMEs (10-249 employees) and large firms (250+ employees). Table 6 presents
the split sample estimation results of employment growth on innovation.
28
Table 6: Employment effects of innovation over the business cycle, SMEs and large firms
Number of employees: 10-249
Dep. var.: EMP
SGR_NEWPD
PC
ORGA
GDPGR
DGP
FGP
Constant
Observations
R2_adjusted
Further tests
Wald-test for β=1 (p-value)
Test on exogeneity
SGR_NEWPD (p-value)
Test on instrument validity
Hansen J- test (p-value)
Diff-in-Hansen test (p-value)
RANGE
COOP
RD
F test on excluded instruments (F)
Test on underidentification: Kleib.-Paap LM (chi2)
Test on weak instruments: Kleib.-Paap F (F)
Inference robust to weak identification: And.-Rub. Wald (chi2)
Upturn
Boom
Downturn
Number of employees: 250+
Recession
Upturn
0.994***
(0.024)
-1.222
(0.792)
-2.065***
(0.449)
3.772***
(0.532)
-1.564*
(0.834)
-1.533
(0.969)
0.972***
(0.032)
-0.350
(1.493)
0.566
(0.703)
2.473
(1.916)
3.159**
(1.253)
1.064
(1.191)
1.005***
(0.025)
-1.795*
(0.977)
-1.553**
(0.648)
-0.655***
(0.188)
0.654
(0.675)
-0.108
(0.779)
0.982***
(0.029)
-0.413
(1.076)
-0.813
(0.515)
0.169
(0.289)
0.763
(0.678)
-2.384***
(0.648)
0.953***
(0.031)
-2.999***
(1.107)
-2.871***
(0.695)
6.020***
(0.556)
-3.393***
(0.963)
-3.209***
(0.995)
-68.663***
(7.064)
55,397
0.386
-30.858*
(16.670)
12,092
0.474
-15.021***
(2.794)
56,859
0.387
3.088*
(1.767)
44,340
0.466
0.806
0.387
0.834
0.000***
0.002***
0.654
Boom
Downturn
Recession
1.030***
(0.042)
0.521
(1.193)
-0.798
(0.749)
-0.949***
(0.206)
-1.731*
(0.934)
-0.419
(1.117)
-82.340***
(6.263)
10,093
0.620
1.013***
(0.049)
-0.175
(2.064)
-1.695
(1.053)
1.129
(2.721)
2.090*
(1.164)
-0.290
(1.424)
-24.198
(24.882)
3,438
0.640
-23.176***
(3.080)
8,582
0.553
0.979***
(0.045)
0.443
(1.214)
-0.839
(0.565)
0.335
(0.419)
-0.986
(0.788)
-3.426***
(0.857)
0.359
(2.271)
6,220
0.557
0.532
0.123
0.798
0.479
0.634
0.000***
0.036**
0.000***
0.003***
0.002***
0.046**
0.465
0.180
0.391
0.189
0.983
0.490
0.373
0.419
0.613
0.357
0.216
0.509
0.460
0.215
0.065*
0.999
0.284
0.992
0.195
0.189
0.189
0.859
0.903
0.961
0.831
0.491
0.258
0.494
0.160
0.587
447.52***
265.775***
530.720***
858.517***
207.5***
61.121***
252.667***
428.026***
312.97***
1223.529***
724.798***
991.793***
283.25***
673.479***
326.223***
324.717***
298.38***
280.652***
460.893***
449.902***
48.29***
43.189***
53.828***
215.935***
125.47***
438.128***
282.665***
258.671***
79.06***
342.047***
122.303***
142.058***
Note: * p<0.1; ** p<0.05; *** p<0.01; weighted IV regressions; clustered standard errors in parentheses. For large firms, estimates for the upturn period are based on the instrument set RANGE and
RD because COOP turned out to be invalid in this regression. For convenience we excluded the estimates of the first stage from this table. Each instrument, however, is positive and highly
significant in the first stage as also indicated by the high F statistics.
29
For both SMEs and large enterprises, higher sales growth rates due to new products are associated
with significantly higher employment growth rates in all phases of a business cycle. The effect slightly
differs between the phases but none of the coefficients is significantly different from one. This
supports our hypothesis H1a for SMEs as well as for large firms. In the previous section we found that
process innovations displace labor in upturns and downturns. Results in Table 6 show that the finding
for downturns is primarily driven by SMEs, whereas process innovation-induced labor displacement
effects in upturns are driven by large enterprises. This implies that there are different dynamics
between SMEs and large firms regarding the labor destruction of process innovations. There are no
significant labor-destructing effects in all other phases. Overall, we do not find support for H2a.
For SMEs, the findings for organizational innovations are in line with the results for the main sample
(sample 1). This supports our hypothesis H3a. The exception is again the recession period for which
we find a negative though insignificant effect. For large firms, we only confirm significant labordestructing effects of organizational innovation during upturn phases.
5.2.2.
Employment decomposition
Table 7 presents the decomposition results for the different size classes. Average employment growth
exhibits a pro-cyclical pattern for both SMEs and large enterprises. In this respect, it is particularly
intriguing that large firms on average have only created employment during boom phases. In contrast,
average employment growth is positive for SMEs in all phases of the business cycle except for the
recession. The decomposition for SMEs is very similar to the one for the total sample presented in
Figure 3, which is not surprising given their weight in firm population. That is, product innovation is
an important driver of employment growth in SMEs. The net employment effect of product innovation
is pro-cyclical and positive in upturn, boom and downturn periods. However, in all three business
cycle phases, the largest contribution to average employment growth in SMEs stems from output
growth for existing products of non-product innovators. This pattern reverses in the recession.
Whereas SMEs without product innovations experience a strong decline in their demand for existing
products, SMEs introducing new products yield new product sales that have almost been sufficiently
high to compensate for the loss in demand of old products. This creates stronger resilience of productinnovating SMEs in recession periods. Labor-destructing effects of process and organizational
innovations are small in SMEs, in particular compared to those generated by the general productivity
trend. The recession period is again an exception in which the general productivity trend indicates
substantial labor hoarding in SMEs.
30
Table 7: Contribution of innovation to employment growth for SMEs and large firms, in %
250+ employees
10-249 employees
Size
Upturn
Boom
Downturn
Recession
4.8
7.4
2.8
-4.8
Employment growth - decomposed into:
-2.5
-0.8
5.4
2.6
0.5
1.6
0.7
2.6
-5.8
8.5
-3.0
-5.6
0.2
8.0
4.6
0.7
1.8
0.9
4.8
-6.1
10.9
3.7
-2.9
-0.2
3.9
2.1
0.4
0.8
0.6
2.5
-6.5
9.0
-2.7
7.3
-0.3
-9.5
-7.3
-0.6
-1.1
-0.5
-2.3
-10.6
8.2
-8.5
General productivity trend in production of old products
-6.9
-7.1
-8.1
3.7
Contribution of process and organizational innovation
-2.0
-1.1
-0.4
-0.5
Output growth due to old products for non-product innovators
thereof for non-innovators
2.2
0.1
3.0
1.2
1.3
0.3
-5.6
-3.2
thereof for process innovators not being organizational innovators
0.2
0.2
0.2
-0.6
thereof for organizational innovators not being process innovators
1.6
0.9
0.4
-1.1
thereof for organisational and process innovators
0.4
0.7
0.4
-0.8
thereof output reduction in old products
3.6
-12.0
8.9
-11.6
4.5
-13.9
-6.1
-20.1
thereof output increase due to new products
15.6
20.6
18.3
13.9
Components
Employment growth - decomposed into:
General productivity trend in production of old products
Contribution of process and organizational innovation
Output growth due to old products for non-product innovators
thereof for non-innovators
thereof for process innovators not being organizational innovators
thereof for organizational innovators not being process innovators
thereof for organisational and process innovators
Net contribution of product innovation for product innovators
thereof output reduction in old products
thereof output increase due to new products
Net contribution of product innovation for product innovators
Note: Decomposition of the weighted-average two-year employment growth rate; the decomposition is based on the
regression results presented in Table 6.
However, the decomposition analysis reveals a very different pattern for large firms. First, large firms
experienced much higher employment losses than SMEs due to improvements in productivity during
upturn, boom and downturn periods. Large firms realized efficiency gains of about 7 to 8% primarily
as a result of the general productivity trend, i.e. selling unprofitable business units, mergers and
acquisitions, learning effects or alike. Displacement effects of organizational and process innovationinduced productivity gains are much smaller, but still larger than in SMEs. Second, larger employment
gains due to product innovation counteract the larger efficiency gains in large firms. Like for SMEs
we find the net employment effect of product innovation to be pro-cyclical and positive in upturn,
boom and downturn periods. In contrast to SMEs, the contribution of product innovation to average
employment growth exceeds by far the one stemming from output growth for existing products of
non-product innovators. Furthermore, employment creation due to demand growth for new and old
products is not sufficiently high to outweigh employment losses due to efficiency gains in upturn and
downturn periods. Third, the most striking result concerns the recession period. In contrast to SMEs,
among the group of large firms non-product innovators perform better in terms of employment growth
than product innovators (-5.6% vs. -6.1%). That is, we do not find large firms to be more resilient to
economic crises. One explanation could be that large firms possess comparative advantages generating
process innovations whereas small firms benefit relatively more from product innovation (Cohen and
Klepper, 1996). Hence, large firms have a stronger incentive to invest in process innovation. The focus
31
on process innovation relative to product innovation might be even stronger in recession periods due
to lower product demand and increased competition based on costs and prices (Spiegel and Stahl,
2014).
6. Conclusion
This paper is the first to examine the effects of different types of innovations on employment growth
over the business cycle. A special focus is devoted to the question whether innovations can be
regarded as a means to become more resilient to economic crises. Despite the literature argues that
employment impacts of innovation crucially depend on product demand-driven effects, there is no
firm-level evidence on this question yet. We analyze the labor-creating and labor-destructing effects of
product, process and organizational innovations over the business cycle using the structural model
developed by Harrison et al. (2014). We estimate the model with CIS firm-level data from 26
European countries covering the period from 1998 to 2010. In total, our sample includes more than
200,000 firm-year observations of the manufacturing sector. To account for business cycle
heterogeneity in the effects of innovation on employment, we split our estimations into the four phases
(upturn, boom, downturn, recession) of a business cycle based on country-specific GDP growth rates.
Descriptive statistics already reveal an interesting pattern: Average employment growth is pro-cyclical
for each type of innovators and for non-innovators. However, employment grows much faster for
innovators than for non-innovators in each phase of a business cycle. This gap in employment growth
between innovators and non-innovators widens during downturn and recession periods.
To estimate the employment growth effects of innovation, we rely on IV estimations using a sensible
identification strategy to correct for potential measurement error. In addition, we quantify the actual
contributions of different types of innovators and non-innovators to average employment growth by
performing a decomposition analysis. Our empirical analysis reveals four major findings. First, the net
employment growth effect of product innovators is strictly pro-cyclical. This effect is positive during
upturn, boom and downturn periods and indicates the labor-creating effect to exceed the labordestructing effect of product innovations. The reverse pattern emerges in recessions in which the net
employment growth effect turns negative, indicating average labor-destruction. Second, product
innovators are more resilient to recessions than non-product innovators. Indeed, when facing negative
economic growth, product innovators on average cut jobs. However, the level of job destruction is
only modest compared to the job destruction in firms that have not implemented any product
innovations between t and t-2. Third, we find resilience only in SMEs but not in large firms. Fourth,
on average, process and organizational innovations particularly reduce the innovators’ labor demand
during upturn and downturn periods. That is, the innovators implemented new processes and business
practices, respectively, that enhanced the production efficiency at the cost of reduced labor demand.
32
Our results have important policy implications. First, we show the strong and positive linkage between
innovation and employment which supports a stronger alignment between these two policy fields.
Second, the employment-preserving effect of product innovations during recessions is a strong
argument for a counter-cyclical public support of product innovation activities. Increasing the public
funding for innovation activities during a recession may help firms to stabilize expectations and
overcome potential financing constraints. However, the precise timing of such measures can be
cumbersome. Automatic stabilizers in public support schemes such as an automatic increase in the tax
credit for R&D in a recession may be an answer to the timing problem. In times of a recession, firms
may have already downsized their R&D personnel or other innovation-related expenditures before a
government adopts and actually pays the counter-cyclical innovation support.
Our study is subject to some limitations. First, the lack of firm identifiers does not allow us to analyze
employment growth effects for the same firm over time. However, we can split the observed firms into
the respective business cycle phases. This allows us to analyze the employment growth dynamics for
different types of innovators and non-innovators for specific growth periods. Second, the correct
identification of the gross effect of process and organizational innovation actually depends on the
availability of data on firm-level price changes of old products. This type of data is generally not
available in firm-level data sets. Therefore, we use industry-level price growth rates as proxy variable.
Larger deviations between firm- and industry-level prices may bias our results. Third, due to lack of
firm-level price data we are also not able to estimate the net employment effect of process and
organizational innovations. That is, we only get an estimate for the labor destruction due to
productivity enhancements but not for the potentially induced labor creation due to price reductions.
Fourth, authors have argued that organizational innovation is a fuzzy concept and difficult to define
(Lam, 2005; Armbruster et al., 2008). Hence, organizational innovation is more likely to be subject to
potential mismeasurement. We use data from the harmonized CIS that includes clearly defined
questions about organizational innovations. Still, survey respondents may be less sure about the
meaning of “organizational innovations” compared to product or process innovation.
Despite the limitations, our analysis draws a positive picture of the ability of innovation to create new
employment. This puts our results in some contrast to other recent research, which points to potential
negative effects of new process and automatization technologies on employment (see e.g. Frey and
Osborne, 2013; Brynjolfsson and McAfee, 2014). It may be that potential losses from innovation are
more visible than the potential benefits of new technologies from today’s perspective. There may also
be a tendency to underestimate benefits and overestimate losses from technological change. The most
important lesson policymakers can learn from our results is that innovations create new jobs
particularly during recessions.
33
References
Acemoglou, D., 2002. Technical change, inequality, and the labour market. Journal of Economic
Literature 40(1), 7-72.
Addison, J.T., Bellmann, L., Schank, T., Teixeira, P., 2008. The demand for labour: an analysis using
matched employer-employee data from the German LIAB. Will the high unskilled worker ownwage elasticity please stand up? Journal of Labour Research 29(2), 114–137.
Aghion, P., Angeletos, G.M., Banerjee, A., Manova, K., 2010. Volatility and growth: Credit
constraints and the composition of investment. Journal of Monetary Economics 57(3), 246-265.
Aghion, P., Ashkenazy P., Berman, N., Eymard, L., Cette, G., 2012. Credit constraints and the
cyclicality of R&D investment: Evidence from France. Journal of the European Economic
Association 10(5), 1001-1024.
Aghion, P., Saint-Paul, G., 1998. Virtues of bad times: Interaction between productivity growth and
economic fluctuations. Macroeconomic Dynamics 2(3), 322-344.
Archibugi, D., Filippetti, A., Frenz, M., 2013. Economic crisis and innovation: Is destruction
prevailing over accumulation? Research Policy 42(2), 303-314.
Armbruster, H., Bikfalvi, A., Kinkel, S., Lay, G., 2008. Organisational innovation: The challenge of
measuring non-technical innovation in large-scale surveys. Technovation 28(10), 644-657.
Autor, D.H., Levy, F., Murnane, R.J., 2003. The skill content of recent technological change: An
empirical exploration. The Quarterly Journal of Economics 118(4), 1279-1334.
Axarloglou, K., 2003. The cyclicality of new product introductions. Journal of Business 76(1), 29-48.
Balleer, A., Gehrke, B., Lechthaler, W., Merkl, C., 2016. Does short-time work save jobs? A business
cycle analysis. European Economic Review 84, 99-122.
Barlevy, G., 2007. On the cyclicality of research and development. American Economic Review 97(4),
1131-1164.
Bauer, T.K., Bender, S., 2004. Technological change, organisational change, and job turnover. Labour
Economics 11(3), 265-291.
Bean, C.R., 1990. Endogenous growth and the procyclical behaviour of productivity. European
Economic Review 34(2), 355-363.
Bernard, A.B., Redding, S.J., Schott, P.K., 2010. Multiple-product firms and product switching.
American Economic Review 100(1), 70-97.
Benavente, J.M., Lauterbach, R., 2007. The effect of innovation on employment, evidence from
Chilean firms, UNU-MERIT Working Paper, Maastricht.
34
Bhaumik, S.K., 2011. Productivity and the economic cycle. BIS Economics Paper No. 12, London.
Bilbiie, F.O., Ghironi, F., Melitz, M.J., 2012. Endogenous entry, product variety, and business cycles.
Journal of Political Economy 120(2), 304-345.
Black, S.E., Lynch, L., 2004. What’s driving the new economy?: The benefits of workplace
innovation. The Economic Journal 114(493), F97-F116.
Blechinger, D., Kleinknecht, A., Licht, G., Pfeiffer, F., 1999. The impact of innovation on employment
in Europe: An analysis using CIS data. ZEW-Dokumentation No. 98-02 218, Mannheim.
Boeri, T., Brücker, H., 2011. Short-time work benefits revisited: some lessons from the great
recession. Economic Policy 26(88), 697-765.
Bresnahan, T.F., Brynjolfsson, E., Hitt, L.M., 2002. Information technology, workplace organisation
and the demand for skilled labour: Firm-level evidence. Quarterly Journal of Economics 117(1),
339-376.
Brynjolfsson, E., McAfee, A., 2014. The second machine age: work, progress, and prosperity in a
time of brilliant technologies. WW Norton & Company, 2014.
Broda, C., Weinstein, D.E., 2010. Product creation and destruction: evidence and price implications.
American Economic Review 100(3), 691-723.
Brouwer, E., Kleinknecht, A., Reijnen, J.O.N., 1993. Employment growth and innovation at the firm
level – an empirical study. Journal of Evolutionary Economics 3(2), 153-159.
Caroli, E., Van Reenen, J., 2001. Skill-biased organisational change? Evidence from a panel of British
and French establishments. The Quarterly Journal of Economics 116(4), 1449-1492.
Cohen, W.M., 1995. Empirical studies of innovative activity. In: Stoneman, P. (Ed.), Handbook of
innovation and technological change. Blackwell, Oxford, 182-264.
Cohen, W.M., 2010. Fifty years of empirical studies of innovative activity and performance. In: Hall,
B.A., Rosenberg, N. (Eds.), Handbook of economics of innovation. Elsevier, Amsterdam, 129213.
Cohen, W.M., Klepper, S., 1996. A Reprise of Size and R&D. The Economic Journal 106(437), 925951.
Corrado, C., Mattey, J., 1997. Capacity utilization. Journal of Economic Perspectives 11(1), 151-167.
Crespi, F., Tacsir, E., 2013. Effects of innovation on employment in Latin America. MERIT Working
Papers No. 2013-001, Maastricht.
Dachs, B., Peters, B., 2014. Innovation, employment growth and foreign ownership of firms. A
European perspective. Research Policy 43(1), 214–232.
35
Damijan, J.P., Kostevc, C., Stare, M., 2014. Impact of innovation on employment and skill upgrading
of firms. Simpatic Working Paper No. 7.
Delgado, M., Porter, M.E., Stern, S., 2015, Clusters and the Great Recession. Paper presented at the
2015 DRUID Conference. Rome:
EC, 2012. Europe’s growth strategy.
Edquist, C., Hommen, L., McKelvey, M.D., 2001. Innovation and employment: Process versus
product innovation. Edward Elgar, Cheltenham.
Entorf, H., Pohlmeier, W., 1990. Employment, innovation and export activity: Evidence from firmlevel data. In: Florens, J.-P., Ivaldi, M., Laffont, J.-J., Laisney, F. (Eds.), Microeconometrics:
surveys and applications. Oxford, 394–415.
Evangelista, R., Vezzani, A., 2011. The impact of technological and organisational innovations on
employment in European firms. Industrial and Corporate Change 21(4), 871-899.
Fabrizio, K.R., Tsolmon, U., 2014. An empirical examination of the procyclicality of R&D investment
and innovation. The Review of Economics and Statistics 96(4), 662-675.
Fagnart, J.F., Licandro, O., Portier, F., 1999. Firm heterogeneity, capacity utilization, and the business
cycle. Review of Economic Dynamics 2(2), 433-455.
Falk, M., 2001. Organisational change, new information and communication technologies and the
demand for labour in services. ZEW Discussion Paper No. 01-25, Mannheim.
Francois, P., Lloyd-Ellis, H., 2003. Animal spirits through creative destruction. American Economic
Review 93(3), 530–550.
Freeman, C., Clark, J., Soete, J., 1982. Unemployment and technical innovation. Pinter, London.
Frey, C.B., Osborne, M.A., 2013. The future of employment: how susceptible are jobs to
computerisation. Retrieved September 7.
Gali, J., Hammour, M.L., 1991. Long run effects of business cycles. Discussion Papers 1991_18,
Columbia University, Department of Economics.
Garcia, A., Jaumandreu, J., Rodriguez, C., 2004. Innovation and jobs: Evidence from manufacturing
firms. MPRA Paper No. 1204, Munich.
Gera, S., Gu, W., 2004. The effect of organisational innovation and information and communications
technology on firm performance. International Productivity Monitor 9, 37-51.
Goos, M., Manning, A., Salomons, A., 2014. Explaining job polarization: Routine-biased
technological change and offshoring. American Economic Review 104(8), 2509-2526.
36
Greenan, N., 2003. Organisational change, technology, employment and skills: An empirical study of
French manufacturing. Cambridge Journal of Economics 27(2), 287-316.
Hall, B.H., Lotti, F., Mairesse, J., 2008. Employment, innovation, and productivity: Evidence from
Italian microdata. Industrial and Corporate Change 17(4), 813-839.
Haltiwanger, J., Jarmin, R.S., Miranda, J., 2013. Who creates jobs? Small versus large versus young.
The Review of Economics and Statistics 95(2), 347–361.
Harrison, R., Jaumandreu, J., Mairesse, J., Peters, B., 2008. Does innovation stimulate employment? A
firm-level analysis using comparable micro-data from four European countries. NBER Working
Paper No. 14216, Cambridge, MA.
Harrison, R., Jaumandreu, J., Mairesse, J., Peters, B., 2014. Does innovation stimulate employment? A
firm-level analysis using comparable micro-data from four European countries. International
Journal of Industrial Organization 35, 29-43.
Himmelberg, C., Petersen, B., 1994. R&D and internal finance: A panel study of small firms in hightech industries. The Review of Economics and Statistics 76(1), 38-51.
Ichniowski, C., Kochan, T.A., Levine, D., Olson, C., Strauss, G., 1996. What works at work: overview
and assessment. Industrial Relations 35(3), 299-333.
Ichinowski, C., Shaw, K., Prennushi, G., 1997. The effects of human resource management practices
on productivity: A study of steel finishing lines. American Economic Review 87(3), 291-313.
Jovanovic, B., 1982. Selection and the evolution of industry. Econometrica 50(3), 649-670.
Kleibergen, F., Paap, R., 2006. Generalized reduced rank tests using the singular value decomposition.
Journal of Econometrics 133(1), 97-126.
Kleinknecht, A., 1989. Firm size and innovation. Small Business Economics 1(3), 215-222.
König, H., Licht, G., Buscher, H., 1995. Employment, investment and innovation at the firm level. In:
OECD (Ed.), The OECD jobs study – investment, productivity and employment. Paris, 67–81.
Lachenmaier, S., Rottmann, H., 2011. Effects of innovation on employment: A dynamic panel
analysis. International Journal of Industrial Organization 29(2), 210-220.
Lam, A., 2005. Organisational innovation. In: Fagerberg, J., Mowery, D.C., Nelson, R.R. (Eds.), The
Oxford handbook of innovation. Oxford University Press, Oxford, 115-147.
Leitner, S.M., Pöschl, J., Stehrer, R., 2011. Changes beget change: Employment effects of
technological and non-technological innovations – a comparison across countries. WIIW
Working Paper No. 72, Vienna.
Leitner, S.M., Stehrer, R., 2012. Labour hoarding during the crisis: evidence for selected new member
states from the financial crisis survey. WIIW Working Paper No. 84, Vienna.
37
Lundvall, B.A., Kristensen, S., 1997. Organisational change, innovation and human resource
development as a response to increased competition. DRUID Working Paper No. 97-16.
Mairesse, J., Wu, Y., Zhao, Y., Zhen, F., 2011. Employment growth and innovation in China: A firmlevel comparison across regions, industries, ownership types and size classes, mimeo, CREST.
Mansfield, E., 1962. Entry, Gibrat’s law, innovation, and the growth of firms. American Economic
Review 52(5), 1023-1051.
Meghir, C., Ryan, A., Van Reenen, J., 1996. Job creation, technological innovation and adjustment
costs: Evidence from a panel of British firms. Annales d’Économie et de Statistique 41/42, 255274.
Mulkay, B., Hall, B., Mairesse, J., 2001. Investment and R&D in France and the United States. In:
Herrmann, H., Strauch, R. (Eds.), Investing today for the world of tomorrow. Springer, New
York, 229-273.
Neumark, D., Wall, B., Zhang, J., 2011. Do small businesses create more jobs? New evidence for the
United States from the National Establishment Time Series. The Review of Economics and
Statistics 93(1), 16-29.
OECD and Eurostat, 2005. Oslo-Manual. Proposed guidelines for collecting and interpreting
innovation data. 3rd Edition. OECD, Paris.
OECD, 2013. OECD employment outlook 2013, Paris.
Ouyang, M., 2011. On the cyclicality of R&D. The Review of Economics and Statistics 93(2),
542-553.
Paunov, C., 2012. The global crisis and firms’ investments in innovation. Research Policy 41(1), 2435.
Peters, B. (2008). Innovation and firm performance an empirical investigation for German firms.
ZEW Economic Studies No. 38, Mannheim.
Peters, B., Riley, R., Siedschlag, I., 2013. The influence of technological and non-technological
innovation on employment growth in European service firms. Servicegap Discussion Paper No.
40, Mannheim.
Petit, P., 1995. Employment and technological change. In: Stoneman, P. (Ed.), Handbook of
innovation and technological change. Blackwell, Oxford.
Pianta, M., 2005. Innovation and employment. In: Fagerberg, J., Movery, D.C., Nelson, R.R. (Eds.),
The Oxford handbook of innovation. Oxford University Press, Oxford, 568-598.
Pianta, M., Lucchese, M., 2012. Innovation and employment cycles. WP-EMS No. 3, Urbino.
38
Piva, M., Santarelli, E., Vivarelli, M., 2005. The skill bias effect of technological and organisational
change: Evidence and policy implications. Research Policy 34(2), 141-157.
Rammer, C., 2012. Schwerpunktbericht zur Innovationserhebung 2010. ZEW-Dokumentation No. 1203, Mannheim.
Rosenberg, N., 1990. Why do firms do basic research (with their own money)? Research Policy 9(2),
165-174.
Schumpeter, J., 1934. The theory of economic development: An inquiry into profits, capital, credit,
interest and the business cycle. Transaction Publishers, New Brunswick and London.
Shleifer, A., 1986. Implementation cycles. Journal of Political Economy 94(6), 1163-1190.
Smolny, W., 2002. Employment adjustment at the firm level. A theoretical model and an empirical
investigation for West German manufacturing firms. Labour 16(1), 65-88.
Spiegel, Y., Stahl, K., 2014. Industry structure and pricing over the business cycle. ZEW Discussion
Paper No. 14-039, Mannheim.
Tether, B.S., Tajar, A., 2008. The organisational-cooperation mode of innovation and its prominence
amongst European service firms. Research Policy 37(4), 720-739.
Thurik, Roy (2009): Entreprenomics: Entrepreneurship, economic growth and policy. In Audretsch,
D.B., Strom, R. (Eds.), Entrepreneurship, growth and public policy. Cambridge University Press.
Van Reenen, J., 1997. Employment and technological innovation: Evidence from U.K. manufacturing
firms. Journal of Labour Economics 15(2), 255-284.
Vivarelli, M., Pianta, M., 2000. The employment impact of innovation: Evidence and policy. London.
Vivarelli, M., 2014. Innovation, employment and skills in advanced and developing countries: A
survey of economic literature. Journal of Economic Issues, 48(1), 123-154.
39
A. Appendix
Table A-1: Country coverage and distribution of the CIS survey waves
Country
Belgium
Bulgaria
Cyprus
Czech Republic
Germany
Denmark
Estonia
Spain
Finland
France
Greece
Croatia
Hungary
Iceland
Italy
Lithuania
Luxembourg
Latvia
Malta
Netherlands
Norway
Portugal
Romania
Sweden
Slovenia
Slovakia
Total
Country
BE
BG
CY
CZ
DE
DK
EE
ES
FI
FR
GR
HR
HU
IS
IT
LT
LU
LV
MT
NL
NO
PT
RO
SE
SI
SK
Wave
1
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
21
2
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
18
3
+
+
+
(+)
(+)
(+)
(+)
(+)
(+)
(+)
(+)
(+)
(+)
+
(+)
4 (15)
4
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
18
5
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
20
Manufacturing
N
%
652
0.3
26,716
13.2
1,217
0.6
11,722
5.8
5,501
2.7
748
0.4
3,631
1.8
38,816
19.2
900
0.4
26,559
13.2
1,318
0.7
1,212
0.6
7,145
3.5
315
0.2
25,929
12.9
2,564
1.3
581
0.3
2,201
1.1
224
0.1
3,445
1.7
3,883
1.9
9,576
4.7
18,477
9.2
3,035
1.5
2,022
1.0
3,302
1.6
201,691
100
Source: CIS3, CIS4, CIS2006, CIS2008 and CIS2010, Eurostat; own calculation.
Note: Values in parentheses reflect that the data for the respective country is available but not the organizational innovation
indicator.
40
Table A-2: Distribution of CIS sample by industry
Industry
Variable
Food / beverages / tobacco
Textile / wearing apparel / leather
Wood / paper / printing
Chemicals
Rubber / plastics
Non-metallic mineral products
Basic and fabricated metals
Machinery
Electrical engineering
Vehicles
N.e.c.
Total
FOOD
TEXT
WOOD
CHEM
PLAS
NONM
BASM
MACH
ELEC
VEHI
NEC
Nace Rev. 1.1.
15-16
17-19
20-21, 22.2-22.3
24
25
26
27-28
29, 33.3
30-32, 33.2, 33.4-33.5
34-35
36, 33.1
Nace 2
N
28,860
28,937
22,820
10,698
11,021
11,493
28,000
20,518
14,973
9,548
14,823
201,691
10-12
13-15
16-18
20-21
22
23
24-25
28, 33
26-27
29-30
31-32
Total
%
14.31
14.35
11.31
5.30
5.46
5.70
13.88
10.17
7.42
4.73
7.35
100
Source: CIS3, CIS4, CIS2006, CIS2008 and CIS2010, Eurostat; own calculation.
Note: Until CIS2006 the industry classification was based on NACE Revision 1.1, since CIS2008 NACE Revision 2 has been
used as industry classification system.
Table A-3: Descriptive statistics on firm characteristics
Firm groups
Total
%
50.5
27.0
11.1
32.1
77.0
18.8
4.2
81.6
13.0
5.4
Non-innovators
Product innovators
Process innovators
Organisational innovators
Small firms
Medium-sized firms
Large firms
Firms pertaining to a domestic enterprise group
Firms pertaining to a foreign enterprise group
Firms not pertaining to an enterprise group
Source: CIS3, CIS4, CIS2006, CIS2008 and CIS2010, Eurostat; own calculation.
Note: Weighted statistics; process innovators refer to firms that implemented only process innovations (PC) during t and t-2.
Table A-4: Distribution of the CIS sample by business cycle phase
Observation period
1998-2000
2002-2004
2004-2006
2006-2008
2008-2010
Total
in %
Upturn
23,756
41,767
0
0
1,945
67,468
33.45
Business cycle phases
Boom
Downturn
12,420
3,868
785
845
2,434
10,682
224
51,784
0
0
15,863
67,179
7.87
33.31
Source: CIS3, CIS4, CIS2006, CIS2008 and CIS2010, Eurostat; own calculation.
41
Recession
0
0
0
0
51,181
51,181
25.38
Table A-5: Sales growth due to new and old products by innovator and business cycle, in %
Firm type
All firms
Innovators
Non-innovators
Process innovators
Product innovators
Organizational innovators
Business cycle phase
Upturn
Boom
Downturn
Recession
Upturn
Boom
Downturn
Recession
Upturn
Boom
Downturn
Recession
Upturn
Boom
Downturn
Recession
Upturn
Boom
Downturn
Recession
Upturn
Total
11.2
16.4
14.1
-10.4
13.6
17.3
16.4
-8.2
8.9
15.0
12.2
-12.5
14.0
17.4
16.3
-9.1
14.8
17.8
16.4
-7.1
14.1
Old products
2.4
4.4
4.8
-19.0
-4.4
-2.4
-4.0
-25.9
8.9
15.0
12.2
-12.5
14.0
17.4
16.3
-9.1
-22.7
-15.3
-19.7
-37.0
-2.1
New products
8.8
12.1
9.3
8.6
18.0
19.7
20.4
17.7
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
37.5
33.1
36.1
29.9
16.2
Boom
17.8
-1.2
19.0
Downturn
17.7
-1.6
19.4
Recession
-8.0
-24.4
16.4
Source: CIS3, CIS4, CIS2006, CIS2008 and CIS2010, Eurostat; own calculation.
Note: Weighted statistics; non-innovators did not introduce product, process or organizational innovations during t and t-2;
“Process innovators” refers to pure process innovators (PC).
42